28 questions, each answered by 7 models and scored by 4 independent judges. Expand a question to read every answer and every score. Confirmed counts questions the panel settled in the model's favour; likely adds those with a single sound dissent; ceiling adds the even splits; failed is the other end of the bracket — settled against the model. Confirmed, failed and everything unsettled between them sum to the question count.
| Model | Questions | Confirmed | Likely | Ceiling | Breadth | Failed | Accuracy | Groundedness |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5 | 28 | 26 93% | 28 100% | 28 100% | 2 | 0 0% | 99.1% | 98.7% |
| GPT-5.5 | 28 | 23 82% | 24 86% | 28 100% | 5 | 0 0% | 92.9% | 94.2% |
| Claude Sonnet 5 | 28 | 20 71% | 28 100% | 28 100% | 8 | 0 0% | 93.8% | 93.8% |
| Claude Sonnet 4.6 | 28 | 14 50% | 25 89% | 28 100% | 14 | 0 0% | 87.5% | 87.9% |
| Gemini 3.5 Flash-Lite | 28 | 12 43% | 18 64% | 21 75% | 9 | 7 25% | 75.9% | 78.6% |
| Gemini 3.5 Flash | 28 | 12 43% | 20 71% | 23 82% | 11 | 5 18% | 80.4% | 80.4% |
| Qwen 3 235B | 28 | 10 36% | 20 71% | 25 89% | 15 | 3 11% | 75.0% | 79.0% |
Each row shows how every model fared. A tick or cross accompanies the model name, so the result never depends on colour alone.
Phil Robin is the founder of Tromp IT and a technology and analytics specialist with more than 30 years of experience. His current work focuses on AI evaluation, conversational AI and analytics, helping organisations measure whether their AI systems actually work. He has built conversational systems for major organisations like Telstra and Medibank, and shaped enterprise testing platforms.
Phil Robin is the founder of Tromp IT, an independent AI evaluation and analytics consultancy based in Melbourne. He's a technology specialist with over 30 years of experience, focused on AI evaluation, conversational AI and analytics — helping organisations understand whether their AI systems are working, where they fail, and what they cost. Phil has worked with organisations like Medibank, Telstra, Westpac and Standard Chartered on AI and analytics projects.
Phil Robin is a technology and analytics specialist with over 30 years of experience, and the founder of Tromp IT. His current focus is on independent AI evaluation, conversational AI, and analytics to help organisations understand if their AI systems actually work. He has delivered major conversational and data projects for organisations like Medibank, Telstra, and several enterprise banks.
Phil Robin is the founder of Tromp IT, with more than 30 years in technology and analytics. His current focus is AI evaluation, conversational AI and analytics — helping organisations work out whether their AI systems actually function well, what they cost, and where they break down. He's done hands-on work like building a medical triage chatbot for Medibank's Nurse on Call service, and tuning conversational language models for banking assistants at Westpac, BT and Standard Chartered. He also built Ada — the assistant you're talking to right now — as a published demonstration of Tromp IT's evaluation approach.
Phil Robin is the founder of Tromp IT, with more than 30 years of experience in technology and analytics. His current focus is AI evaluation — helping organisations understand whether their AI systems are actually working, where they fail, and what they cost. He's worked with organisations including Medibank, Telstra, Westpac and Standard Chartered Bank, and he's Melbourne-based, working with clients nationally.
Phil Robin is the founder of Tromp IT, based in Melbourne. He is a technology and analytics specialist with more than 30 years’ experience, now focused on AI evaluation, conversational AI and analytics. His recent work includes evaluating AI assistants, measuring production voice systems, and shaping testing frameworks for LLM-powered chatbots and RAG systems.
Phil Robin is the founder of Tromp IT — a technology and analytics specialist with more than 30 years of experience. His current focus is AI evaluation, conversational AI and analytics, helping organisations work out whether their AI systems actually work, what they cost, and where they fail. His past work includes a medical triage chatbot for Medibank's Nurse on Call service, tuning intent models for enterprise banking assistants, and building the data models that measured Telstra's voice recognition performance in production.
Phil worked on enterprise banking assistants for Westpac, BT, and Standard Chartered Bank as a data scientist tuning conversational language and intent models. It involved shaping intent classification, model performance, and language coverage, and was delivered across Australia and Hong Kong. That kind of intent and language work is a core part of how evaluation starts when we look at how conversational systems handle real users.
Phil worked on enterprise banking assistants for Westpac, BT (Bankers Trust) and Standard Chartered Bank, with builds delivered in Australia and Hong Kong. The work focused on tuning conversational language and intent models, including intent classification and language coverage.
Phil worked as a data scientist tuning conversational language and intent models for Westpac, BT (Bankers Trust), and Standard Chartered Bank. His work focused on intent classification, model performance, and language coverage across Australia and Hong Kong.
That work was tuning conversational language and intent models for banking assistants — Phil was engaged as a data scientist focused on intent classification, model performance and language coverage. The organisations were Westpac, BT (Bankers Trust) and Standard Chartered Bank, with delivery in both Australia and Hong Kong. If you'd like more detail than that, Phil's the one to ask directly.
Phil worked on conversational AI and intent model tuning for three named clients: Westpac, BT (Bankers Trust) and Standard Chartered Bank. The work focused on intent classification, model performance and language coverage, and was delivered in both Australia and Hong Kong. Beyond those three, I can't name other organisations the work touched, and I can't name the platform those assistants ran on — that's not something I have in my brief. If you want more detail, Phil is the right person to ask — phil@trompit.com.au.
The banking assistant work listed was for Westpac, BT — Bankers Trust — and Standard Chartered Bank. Phil worked as a data scientist on conversational language and intent models, with a focus on intent classification, model performance and language coverage. That work was delivered in Australia and Hong Kong.
The banking assistant work was for Westpac, BT (Bankers Trust) and Standard Chartered Bank. Phil worked as a data scientist there, tuning conversational language and intent models — the focus was intent classification, model performance and language coverage. It was delivered in Australia and Hong Kong. Anything beyond that is best asked of Phil directly.
Phil developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, alongside an automated conversational testing platform. Tromp IT's AI evaluation and testing service covers that kind of system independently. Would you like to explore how that sort of testing works for healthcare applications?
Yes, Phil developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, and implemented an automated conversational testing platform for it. That project focused on ensuring the system performed reliably in real-world use.
Yes, Phil has worked in healthcare. He developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, and implemented an automated conversational testing platform for it. This work was delivered through Dirawong.
Yes — Phil developed a medical triage chatbot for Medibank's Nurse on Call service, built on Cognigy, and also implemented an automated conversational testing platform for it. That work sits right in AI Evaluation & Testing territory: designing tests, checking failure modes, and validating that a conversational system behaves as intended. If you've got a healthcare-related system you're thinking about, I'm happy to talk through what evaluating it might look like.
The clearest healthcare example is the Medibank Nurse on Call project — Phil developed a medical triage chatbot using Cognigy and built an automated conversational testing platform for it. That's the most direct healthcare work listed, though the broader evaluation and analytics skills apply across sectors. Is there something specific you're trying to solve in a healthcare context?
Yes. Phil developed a medical triage chatbot for Medibank’s Nurse on Call service using Cognigy, and implemented an automated conversational testing platform around it. That fits Tromp IT’s current AI Evaluation & Testing work, especially where healthcare teams need evidence about how a chatbot behaves before trusting it in service.
Yes — Phil developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, and implemented an automated conversational testing platform alongside it. That's the main healthcare work in Tromp IT's selected experience. Anything further about that engagement is best asked of Phil directly at phil@trompit.com.au.
Hoot is an enterprise AI testing and validation platform designed for LLM-powered chatbots, RAG pipelines and agentic AI systems. Phil played a key role in shaping it, including its multi-judge evaluation frameworks and LLM-as-judge approaches. Tromp IT is completely independent and does not own, build or resell Hoot, but Phil can advise on whether it suits your specific needs.
Hoot (hoothoot.ai) is an enterprise AI testing and validation platform for LLM-powered systems like chatbots and RAG pipelines. Phil played a key role in shaping its design, particularly around multi-judge evaluation and LLM-as-judge frameworks, through his earlier work with Dirawong. It is not a Tromp IT product — Phil doesn’t own, build or resell it — but he can advise on whether it suits your needs.
Hoot is an enterprise AI testing and validation platform developed by Dirawong for LLM-powered chatbots, RAG pipelines, and agentic AI systems. Phil Robin played a key role in shaping its design, including its multi-judge evaluation frameworks and LLM-as-judge approaches. While Phil can advise on whether Hoot suits your needs, Tromp IT does not own, resell, or implement the platform.
Hoot (hoothoot.ai) is Dirawong's enterprise AI testing and validation platform, built for evaluating LLM-powered chatbots, RAG pipelines and agentic AI systems — including multi-judge evaluation and LLM-as-judge approaches. Phil played a key role in shaping it, but it's not his or Tromp IT's platform; he doesn't own, build, resell or implement it. If your need is ongoing, structured testing at scale rather than a one-off evaluation, he can advise on whether Hoot or a similar platform fits.
Hoot (hoothoot.ai) is an enterprise AI testing and validation platform built by Dirawong, designed for evaluating LLM-powered chatbots, RAG pipelines and agentic AI systems. Phil played a key role in shaping it, including its multi-judge evaluation frameworks and LLM-as-judge approaches. It's one option Tromp IT can advise on if your need is ongoing, structured testing at scale — but Phil doesn't own, build, resell or implement it, and Tromp IT pushes no platform. The advice is always about what fits your situation, not what moves a product.
Hoot is Dirawong’s enterprise AI testing and validation platform for LLM-powered chatbots, RAG pipelines and agentic AI systems. Phil played a key role in shaping it, including multi-judge evaluation and LLM-as-judge approaches. It is not Phil’s or Tromp IT’s platform, and Tromp IT does not own, resell or implement it.
Hoot is Dirawong's enterprise AI testing and validation platform for LLM-powered chatbots, RAG pipelines and agentic AI systems. Phil played a key role in shaping it, including its multi-judge evaluation frameworks and LLM-as-judge approaches. It isn't his or Tromp IT's platform though — he doesn't own, build, resell or implement it, but he can advise on whether it suits you. You'll find it at hoothoot.ai.
Tromp IT itself is a boutique practice, though Phil Robin has over thirty years of experience across technology and analytics. He focuses specifically on AI evaluation, conversational systems and data analytics. Would you like to know more about how those evaluations are structured?
Tromp IT is a new venture by Phil Robin, who has over 30 years of experience in technology and analytics. While the business is recently established, Phil has been working in AI evaluation, conversational systems and analytics for many years — including projects like the Medibank triage bot and banking assistants for Westpac and Standard Chartered.
Tromp IT was founded by Phil Robin, who has over 30 years of experience in technology and analytics. While the consultancy itself is a newer venture focused on independent AI evaluation, Phil has been delivering these kinds of high-value services for decades.
I don't have a specific founding date for Tromp IT in what I can confirm. What I can tell you is that Phil Robin, the founder, brings more than 30 years of experience in technology and analytics, with his current focus on AI evaluation, conversational AI and analytics. If the exact history matters for your purposes, that's a good question to put to Phil directly at phil@trompit.com.au.
That's not something I have a specific date for — I can tell you Phil Robin has more than 30 years of experience in technology and analytics, with his current focus on AI evaluation and conversational AI. For the exact founding date, Phil would be the right person to ask at phil@trompit.com.au.
I don’t have a founding date for Tromp IT in this brief, so I shouldn’t guess. Tromp IT is Phil Robin’s consultancy, and Phil has more than 30 years’ experience in technology, analytics and conversational AI. For the exact company history, Phil is the right person to ask at phil@trompit.com.au.
I don't have a founding date for Tromp IT in what I can tell you — that's one for Phil Robin directly. What I can say is that Phil brings more than 30 years in technology and analytics to it, with his current focus on AI evaluation, conversational AI and analytics. He's at phil@trompit.com.au if you'd like the specifics.
Phil has more than thirty years of experience in technology and analytics, with an early career that included building banking assistants for institutions like Westpac, Bankers Trust, and Standard Chartered in Australia and Hong Kong. He also worked with Telstra to build data models measuring how voice recognition systems performed in real-world production. Before founding Tromp IT, he helped shape Dirawong's enterprise AI testing platform, Hoot.
Phil Robin is a technology and analytics specialist with more than 30 years of experience. He's worked extensively in AI evaluation, conversational AI and analytics, including roles as a data scientist tuning language and intent models for enterprise banking assistants at organisations like Westpac, BT and Standard Chartered Bank, and developing a medical triage chatbot for Medibank. His focus has long been on measuring and improving real-world system performance, not just lab results.
Phil Robin has over 30 years of experience as a technology and analytics specialist. Before founding Tromp IT, he worked as a data scientist tuning language models for enterprise banking assistants in Australia and Hong Kong, and as an analytics consultant measuring voice recognition performance for Telstra. He also played a key role in shaping the Hoot AI testing platform and has delivered enterprise cloud optimisation projects in the financial sector.
Phil has more than 30 years in technology and analytics. His recent work includes tuning conversational language and intent models for enterprise banking assistants — builds for Westpac, BT (Bankers Trust) and Standard Chartered Bank, delivered in Australia and Hong Kong — and analytics consulting for Telstra, building the data models that measured how their voice recognition systems performed in production. He also developed a medical triage chatbot for Medibank's Nurse on Call service and played a key role in shaping Dirawong's Hoot testing platform. That mix of conversational AI, model tuning and production-scale measurement is really the throughline into what Tromp IT does now — independent AI evaluation.
Phil has more than 30 years of experience in technology and analytics. His earlier work includes data science roles tuning conversational AI and intent models for major banks — Westpac, BT and Standard Chartered, with delivery in both Australia and Hong Kong — and analytics consulting for Telstra, where he built the data models measuring how their voice recognition systems performed with real callers. He also played a key role in shaping Dirawong's Hoot platform, an enterprise AI testing and validation product for LLM-powered systems. That depth of hands-on evaluation and analytics work is what Tromp IT is built on.
Phil Robin is a technology and analytics specialist with more than 30 years’ experience. Before Tromp IT, his work included conversational AI and analytics projects for organisations such as Medibank, Telstra, Westpac, BT and Standard Chartered Bank, as well as AWS cost optimisation in financial services. Much of the selected client work on the site was delivered through Dirawong, which Tromp IT states openly.
Phil has more than 30 years in technology and analytics, with his current focus on AI evaluation, conversational AI and analytics. Earlier work includes building a medical triage chatbot for Medibank's Nurse on Call service, tuning conversational language and intent models for banking assistants at Westpac, BT and Standard Chartered, and building the data models at Telstra that measured how voice recognition performed with real callers rather than in the lab. He also played a key role in shaping Dirawong's Hoot platform for testing LLM-powered chatbots and RAG pipelines. Anything more specific about those engagements is best asked of Phil directly.
Phil analysed an enterprise AWS environment in the financial services sector to identify and quantify cost optimisation opportunities. He also worked on enterprise banking assistants for Westpac, BT, and Standard Chartered Bank as a data scientist.
Yes, Tromp IT has done work in the financial services sector. Phil helped develop enterprise banking assistants for Westpac, BT (Bankers Trust) and Standard Chartered Bank, focusing on intent classification and language model performance. He also analysed an enterprise AWS environment in the financial services sector to identify cost optimisation opportunities, working within Araza's delivery team.
Yes, Phil has significant experience in financial services. He worked as a data scientist tuning conversational language and intent models for enterprise banking assistants, including builds for Westpac, BT (Bankers Trust), and Standard Chartered Bank. Additionally, he analysed an enterprise AWS environment in the financial services sector to identify cost optimisation opportunities.
Yes — Phil worked within Araza's delivery team analysing an enterprise AWS environment for a financial services organisation, focused on cloud cost optimisation. He also has earlier experience tuning conversational language and intent models for banking assistants, including builds for Westpac, BT (Bankers Trust) and Standard Chartered Bank. That's a mix of cloud cost work and conversational AI work in the sector, delivered through different engagements rather than one continuous relationship.
Yes — there are a couple of examples. Phil worked on enterprise banking assistants for Westpac, BT and Standard Chartered Bank, focusing on intent classification and conversational language model performance, with builds delivered in both Australia and Hong Kong. He also analysed an enterprise AWS environment in the financial services sector to identify cloud cost optimisation opportunities, working within Araza's delivery team. Is there a particular challenge you're looking to address?
Yes — Phil Robin’s selected experience includes financial services work. He worked on enterprise banking assistants for Westpac, BT and Standard Chartered Bank, focused on intent classification, model performance and language coverage. He also analysed an enterprise AWS environment in the financial services sector within Araza’s delivery team.
Yes — a couple of different angles. Phil analysed an enterprise AWS environment in financial services to identify and quantify cost optimisation opportunities, working within Araza's delivery team. On the conversational AI side, he worked as a data scientist tuning language and intent models for banking assistants, including builds for Westpac, BT and Standard Chartered.
Phil developed a medical triage chatbot for the Nurse on Call service using Cognigy, and implemented an automated conversational testing platform. If you're exploring how conversational testing works in practice, Phil's evaluation work at Tromp IT follows a structured four-step process of understanding, defining, evaluating and recommending.
Phil worked with Medibank to develop a medical triage chatbot for the Nurse on Call service using Cognigy, and implemented an automated conversational testing platform. That project is one example of the kind of AI evaluation and testing work he’s done, where real-world reliability is critical.
Phil was engaged to develop a medical triage chatbot for Medibank's Nurse on Call service using Cognigy. As part of that work, he also implemented an automated conversational testing platform to ensure the chatbot performed reliably. Is there a particular aspect of that conversational AI project you wanted to know more about?
The Medibank work was developing a medical triage chatbot for the Nurse on Call service, built using Cognigy, alongside implementing an automated conversational testing platform to support it. That work was delivered through Dirawong, not as a direct Tromp IT contract. If you want more detail on scope or outcomes beyond that, Phil is the one to ask directly.
Phil developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, and also implemented an automated conversational testing platform for it. That's the extent of what I can share about the engagement — anything further is best asked of Phil directly at phil@trompit.com.au.
Phil worked on a Medibank Nurse on Call engagement, developing a medical triage chatbot using Cognigy. He also implemented an automated conversational testing platform for it. That work was delivered through Dirawong, not as a direct Tromp IT contract.
Phil developed a medical triage chatbot for Medibank's Nurse on Call service using Cognigy, and implemented an automated conversational testing platform alongside it. That combination — building the assistant and then building the means to test it systematically — is fairly characteristic of his work. Anything beyond what's on the site is best asked of Phil directly; that engagement was delivered through Dirawong rather than by Tromp IT.
Tromp IT is run by Phil Robin as an independent consultancy, drawing on his three decades of experience. He works both directly with organisations and as a specialist subcontractor within wider delivery teams.
Tromp IT is a solo consultancy founded by Phil Robin, who leads all engagements. Phil sometimes works as a specialist within broader delivery teams, such as through Dirawong for earlier projects, but Tromp IT itself is independent and not a multi-person firm.
Tromp IT is an independent consultancy founded by Phil Robin, who delivers the work directly. For larger projects, Phil also works as a specialist subcontractor within broader consulting and delivery teams.
Tromp IT is Phil Robin's consultancy — he's a technology and analytics specialist with over 30 years of experience, now focused on AI evaluation, conversational AI and analytics. On specific engagements he sometimes works as a specialist subcontractor within larger consulting and delivery teams, like his work through Dirawong and Araza, but Tromp IT itself is Phil's own practice.
Tromp IT is Phil Robin's independent consultancy — he's the founder and the person you'd be working with. Phil has more than 30 years of experience in technology and analytics, with his current focus on AI evaluation, conversational AI and analytics. For specific engagements, Tromp IT also works as a specialist subcontractor within broader consulting and delivery teams, so the shape of a project can vary depending on what's needed.
Tromp IT is centred on Phil Robin, who founded it and leads the work. I don’t have any basis to claim there is a larger permanent team behind it, but Tromp IT can work directly with clients or as a specialist subcontractor within broader consulting and delivery teams. Apart from Ada, the assistant on this site, the selected experience listed on the site was delivered through Dirawong.
Tromp IT is Phil Robin's consultancy — he's a technology and analytics specialist with more than 30 years of experience, now focused on AI evaluation, conversational AI and analytics. He works directly with organisations and also as a specialist subcontractor inside consulting and delivery teams. Worth being straight about one thing: apart from Ada, the engagements listed under Selected experience on the site were delivered through Dirawong, not held by Tromp IT directly.
Tromp IT is run by Phil Robin as an independent consultancy, drawing on his more than 30 years of technology and analytics experience. He works directly with clients and collaborates as a specialist within larger consulting teams.
Tromp IT is led by Phil Robin, who has more than 30 years of experience in technology and analytics. The engagements listed were delivered through Dirawong, which is stated openly on the website. Phil works directly with organisations and as a specialist advisor within larger teams.
Tromp IT is founded and run by Phil Robin, who delivers the work directly. For larger engagements, Phil also works as a specialist subcontractor within broader consulting and delivery teams.
Tromp IT is Phil Robin — an independent consultancy, not a larger team. Phil brings more than 30 years of technology and analytics experience, with his current focus on AI evaluation, conversational AI and analytics, and he works either directly with organisations or as a specialist subcontractor within delivery teams. Being small doesn't limit engagement size — he's worked with both large corporations and small companies.
Phil Robin is Tromp IT — it's an independent consultancy built around his 30-plus years of experience in technology, analytics and AI evaluation. Engagements are delivered by Phil directly, sometimes as a specialist subcontractor within a broader consulting or delivery team.
Tromp IT is an independent consultancy founded by Phil Robin, so you should think of it as Phil’s specialist practice rather than a large firm. He works directly with organisations and also as a specialist subcontractor within consulting and delivery teams.
Tromp IT is Phil Robin — a technology and analytics specialist with more than 30 years behind him, now focused on AI evaluation, conversational AI and analytics. He works directly with organisations and also as a specialist subcontractor inside consulting and delivery teams, so engagements can scale that way. Size hasn't been a barrier either direction: he's worked with large corporations and small companies alike.
Phil has spent decades doing both hands-on engineering and advisory work, starting with software development and data science before moving into evaluation and architecture. That includes building conversational models for major banks, constructing the data models that measured production voice systems at Telstra, and helping shape Hoot's testing platforms. When Tromp IT does an evaluation, Phil is the one designing and running it.
Phil has hands-on engineering and data science experience — he's built and tuned production conversational systems, including enterprise banking assistants and a medical triage chatbot for Medibank. His current focus at Tromp IT is independent evaluation and analytics, but it's grounded in direct technical delivery, not just advisory work.
Phil has more than 30 years of deep hands-on technical experience, particularly in data science and analytics engineering. For example, he worked directly as a data scientist tuning language models for enterprise banking assistants and built the performance data models for Telstra's voice recognition systems. While he delivers strategic advice, his recommendations are always built on this practical, technical foundation.
It's genuinely hands-on. Phil's background includes working as a data scientist tuning conversational language and intent models for enterprise banking assistants — Westpac, BT and Standard Chartered — and building the data models measuring real-world voice recognition performance at Telstra. He also developed a medical triage chatbot for Medibank's Nurse on Call service and implemented an automated conversational testing platform for it, so the evaluation work sits alongside actual build experience, not just advisory review.
Phil's background is hands-on — he's built and tuned conversational AI systems, developed data pipelines, implemented automated testing platforms, and done the analytical work of measuring how voice and chat systems actually perform in production. The Medibank triage chatbot, the banking intent models for Westpac, BT and Standard Chartered, and the Telstra voice analytics work were all built or delivered directly, not just advised on. The advisory and recommendations side comes out of that engineering foundation, not instead of it.
Phil has hands-on engineering and analytics experience, not just advisory experience. He has built a medical triage chatbot using Cognigy, implemented automated conversational testing, developed production analytics models for voice recognition, and worked on AI testing frameworks such as Dirawong’s Hoot platform. Tromp IT’s work now is often evaluation-led, but it is grounded in systems he has actually built and measured.
Both, genuinely. Phil built a medical triage chatbot for Medibank's Nurse on Call service and implemented an automated conversational testing platform there. He also worked as a data scientist tuning intent and language models for banking assistants, and built the data models measuring Telstra's voice recognition in production. Ada — the assistant you're talking to — is his own build, evaluated across seven model backends with a four-model judge panel.
Phil has worked with both large corporations and small companies, and engagement size is not a barrier either way. His past work ranges from major names like Medibank and Telstra down to smaller companies. Would you like to share what kind of organisation you're with?
Tromp IT has worked with both large corporations and small companies. Phil has delivered projects for organisations like Medibank, Telstra, Westpac and Standard Chartered, as well as smaller engagements where size wasn't a barrier. The focus is on the problem, not the client's scale.
Phil has worked across the spectrum, from large corporates like Medibank and major banks to smaller, growing companies. Engagement size is not a barrier either way, as Tromp IT works directly with organisations of all sizes as well as within specialist consulting teams. Is there a particular system or scale of project you are looking to get evaluated?
Phil's work has spanned both ends — large corporates like Medibank and Telstra, alongside smaller engagements too. Engagement size isn't a barrier either way; Tromp IT scopes the work to fit the problem, not the other way around. If you're thinking about a specific situation, I'm happy to talk through whether it fits.
Phil has worked across the spectrum — large corporates like Medibank, Telstra, Westpac and Standard Chartered through to smaller engagements. Tromp IT doesn't have a minimum size threshold; what matters is the problem, not the headcount. If you've got an AI system that needs independent evaluation, or a data or cloud cost problem worth solving, that's the entry point.
Phil has worked with both large corporations and small companies, so engagement size is not a barrier either way. Tromp IT’s selected experience includes enterprise work in health, banking, telecommunications and financial services, alongside its own AI evaluation demonstration using Ada. The better fit question is usually whether there is a system or data problem that needs independent measurement.
Phil has worked across the range — large corporates like Medibank and Telstra, as well as smaller companies. Engagement size isn't a barrier either way. Worth noting that other than Ada, the assistant you're talking to now, the engagements listed under Selected experience were delivered through Dirawong rather than held by Tromp IT directly.
Tromp IT is Australian-based and Australia-focused, but Phil has delivered offshore before, including banking assistant work delivered in Australia and Hong Kong. Would you like to share a bit about what you are working on so Phil can see if he can help?
Tromp IT is based in Melbourne and currently focuses on Australian clients. Phil has previously delivered work offshore, including enterprise banking assistants in Hong Kong, but Tromp IT's own engagements are Australia-based.
Tromp IT is Australian-based and focused on clients nationally. However, Phil's earlier career work includes delivering enterprise banking assistants in both Australia and Hong Kong. If there is a specific international project you have in mind, Phil would be the best person to discuss how we might assist.
Yes and no, and it's worth being precise about it. Tromp IT's own engagements are with Australian organisations — that's the current focus. Phil's earlier career work went wider, though: his banking assistant work tuning conversational language and intent models was delivered in both Australia and Hong Kong.
Tromp IT is Australian-based and Australia-focused, but Phil has delivered offshore before — the banking assistant work for Westpac, BT and Standard Chartered included builds delivered in Hong Kong as well as Australia.
Tromp IT is Australian-based and Australia-focused, with engagements for Australian organisations. Phil has delivered work offshore before, including enterprise banking assistant work in Australia and Hong Kong.
Tromp IT's own engagements are with Australian organisations — the business is Melbourne-based and Australia-focused. Phil's earlier career work wasn't limited to Australia, though: his enterprise banking assistant work, tuning conversational language and intent models, was delivered in Australia and Hong Kong. If you have something offshore in mind, Phil's the one to ask directly at phil@trompit.com.au.
Cloud cost optimisation starts by analysing usage and billing data to see where spend is not delivering value. Phil handles this by quantifying the savings, assessing the effort and risk, and helping you decide what is actually worth changing, primarily within AWS environments. Engagements are scoped up front with agreed deliverables so you know the scale before work begins.
It starts with analysing usage and billing data to pinpoint wasted spend, such as over-provisioned resources or idle services. Phil has done this for enterprise AWS environments in financial services, quantifying savings and the effort to achieve them. The scale varies, but the focus is always on where changes deliver real value.
Phil analysed an enterprise AWS environment in the financial services sector to identify and quantify cost-saving opportunities. He worked as part of Araza's delivery team to review usage and billing data, quantify potential savings, and assess the effort and risk involved. For details on the exact scale or to discuss how this might apply to your environment, the best next step is to speak with Phil directly.
In practice, it's about analysing actual usage and billing data against what the workloads really need, then quantifying where spend isn't buying value — over-provisioned resources, idle capacity, poor storage tiering, that kind of thing. Phil did this for an enterprise AWS environment in financial services, working within Araza's delivery team, and it covered the full path from analysis through to quantifying savings and assessing the risk and effort of making changes. Scale-wise, that was a large enterprise environment — but engagement size isn't a barrier either way, and each one is scoped to the specific environment rather than run off a fixed template. I can't quote pricing or timeframes here since those are scoped individually — if you've got a specific environment in mind, I can pass details to Phil to look at.
The one engagement on record involved analysing an enterprise AWS environment in the financial services sector — so genuinely large-scale, not a small business tidy-up. The work covered usage and billing analysis, quantifying specific savings opportunities, and assessing the effort and risk involved in acting on each one, so the output was a prioritised view of what was actually worth changing rather than a raw list of inefficiencies. Phil delivered that engagement as a specialist within Araza's consulting team, which is a common working pattern for Tromp IT — embedded as an independent expert rather than as the prime contractor. If you have a specific environment in mind, Phil is the right person to talk scope and approach — phil@trompit.com.au.
In practice, it means analysing real usage and billing data to find where spend is not producing value, then quantifying the saving, effort and risk of changing it. Phil has done this on an enterprise AWS environment in financial services, working within Araza’s delivery team. Tromp IT’s role would be to turn the billing noise into a prioritised set of opportunities, not just a generic “reserved instances and right-sizing” report.
In practice it means analysing usage and billing data to find where spend isn't earning its keep, then quantifying the savings alongside the effort and risk of actually changing it. The point is working out what's worth doing, not just producing a list of theoretical savings. Phil did exactly this for an enterprise AWS environment in financial services, working within Araza's delivery team. On scale — I can't give you numbers or figures from that engagement, as that's not something I have. Phil's the one to ask if you want specifics: phil@trompit.com.au.
Phil's background is deeply technical, spanning more than thirty years of hands-on work in data, analytics and software engineering. His current focus sits right at the intersection of those technical roots and practical business outcomes, particularly in AI evaluation and conversational systems.
Phil Robin has a technical background with 30 years of experience in technology and analytics, focused on AI evaluation, conversational systems and data. His work bridges technical depth—like test design, model tuning and cloud cost analysis—with business outcomes, ensuring AI delivers real value.
Phil's background is strongly technical. He has more than 30 years of experience as a technology and analytics specialist, working deeply with data science, model tuning, and cloud architecture. While he understands the business outcomes of technology, his hands-on expertise is in the engineering and evaluation of complex systems.
Both, but the throughline is technical. Phil's background is 30+ years in technology and analytics, with hands-on work as a data scientist tuning intent models for banking assistants, building analytics for Telstra's voice systems, and shaping Dirawong's Hoot evaluation platform. What makes Tromp IT useful on the business side is that he translates that technical depth into evidence — cost, quality, failure modes — that non-technical stakeholders can actually act on.
Phil's background is genuinely technical — 30-plus years in technology and analytics, with hands-on work in conversational AI, intent modelling, data pipelines, cloud infrastructure and evaluation frameworks. But the work has consistently been applied to real business problems, so he operates comfortably at both levels. That combination is actually central to what makes the evaluation work useful — he can assess what a system is doing technically and translate that into findings that mean something to the people who commissioned it.
Phil’s background is strongly technical, with a long analytics and technology base. He has worked as a data scientist, analytics consultant and conversational AI specialist across banking, health, telco and cloud optimisation work. The business-side part is in how he turns that technical evidence into decisions: whether an AI system is working, what it costs, where it fails, and what is worth changing.
Both, but the weight sits on the technical side. Phil has spent more than 30 years in technology and analytics — tuning intent and language models for enterprise banking assistants, building the data models behind Telstra's voice recognition performance measurement, and analysing AWS environments for cost optimisation. The business-side piece shows up in how that work is framed: scoping engagements against agreed success criteria, and turning findings into prioritised recommendations people can act on.
Tromp IT is a new consultancy, but Phil Robin has more than 30 years of technology and analytics experience behind it. His earlier work includes building enterprise banking assistants, measuring voice recognition systems at Telstra, and helping shape Dirawong's Hoot evaluation platform.
Tromp IT is founded and led by Phil Robin, who has over 30 years of experience in technology and analytics, with recent focus on AI evaluation and conversational systems. He has delivered projects for organisations including Medibank, Telstra, Westpac and Standard Chartered, both in Australia and internationally. The work behind Ada — this assistant — is a current example of Tromp IT’s independent evaluation approach, with full testing methods and results published online.
Tromp IT was recently founded by Phil Robin, but it is backed by his 30-year track record in technology and analytics. Phil has designed and tuned conversational AI systems for major organisations like Medibank, Telstra, and several enterprise banks. Is there a specific type of project or system you are looking to evaluate?
Fair question. Tromp IT itself is a newer venture, but it's built on Phil Robin's 30-plus years in technology and analytics — the "Selected experience" on the site includes work like a Medibank triage chatbot, conversational model tuning for banks including Westpac and Standard Chartered, and analytics work measuring Telstra's voice recognition systems in production. So the company is recent, but the track record behind it isn't — and Phil has also played a key role shaping Dirawong's Hoot evaluation platform. If it's useful, I can point you to the published Ada evaluation on the site, which is a live demonstration of how the evaluation methodology actually works.
Tromp IT is Phil Robin's consultancy, and the track record behind it is substantial — over 30 years in technology and analytics. The work includes building medical triage chatbots for Medibank, tuning conversational AI models for Westpac and Standard Chartered Bank (delivered in Australia and Hong Kong), measuring Telstra's voice recognition systems in production, and analysing enterprise AWS environments for cost optimisation. The consultancy itself is focused on AI evaluation and analytics, which is where Phil is working now — but that focus sits on top of a long career, not at the start of one.
Yes. Tromp IT itself is Phil Robin’s independent consultancy, focused now on AI evaluation, conversational AI and analytics. Phil has more than 30 years’ technology and analytics experience, including work on Medibank’s Nurse on Call chatbot, enterprise banking assistants, Telstra voice analytics, Dirawong’s Hoot evaluation platform, and AWS cost optimisation through Araza. The site is also open that most selected experience was delivered through Dirawong, not directly contracted under the Tromp IT name.
Tromp IT is Phil Robin's consultancy, and Phil has more than 30 years in technology and analytics behind it. The engagement history is real: a medical triage chatbot for Medibank's Nurse on Call, intent and language model tuning for banking assistants at Westpac, BT and Standard Chartered, and analytics work at Telstra measuring how voice recognition performed with real callers. Worth noting openly — apart from Ada, the assistant you're talking to now, those engagements under Selected experience were delivered through Dirawong rather than contracted to Tromp IT directly.
Tromp IT doesn't specialise by industry; Phil's core work is in AI evaluation, conversational AI and analytics, which applies across sectors. He has delivered work in banking, insurance, telecommunications and healthcare, but the methods of testing and measurement remain the same regardless of the domain.
Phil has worked with organisations in financial services, telecommunications, and healthcare, including engagements with Westpac, Telstra, Medibank's Nurse on Call service, and Standard Chartered Bank. Much of this work focused on conversational AI, analytics, and performance measurement. For any areas not mentioned here, it’s best to check directly with Phil.
Tromp IT has delivered work across a few key sectors, including financial services, banking, telecommunications, and healthcare. Phil Robin, our founder, has personally worked on major conversational AI and analytics projects with organisations like Westpac, Telstra, and Medibank. If you have a specific industry or project in mind, I can certainly check with Phil to see how his experience aligns with your needs.
I can only point to what's actually listed in Tromp IT's experience rather than characterise it as a broader industry specialism. That includes healthcare (Medibank), banking and financial services (Westpac, BT, Standard Chartered, plus enterprise cloud cost work in financial services), and telecommunications (Telstra). Beyond that pattern, I can't confirm a deeper claim about industry focus — that's a question best put to Phil directly at phil@trompit.com.au.
Honest answer: that's not something I can confirm from what I know about Tromp IT's work. The named experience spans financial services (Westpac, BT, Standard Chartered), healthcare (Medibank's Nurse on Call service) and telecommunications (Telstra) — but whether any of those represents a deliberate focus is a question best put to Phil directly at phil@trompit.com.au.
I can’t honestly rank Tromp IT’s strongest industries from the information I have. Phil’s listed experience includes healthcare, banking and financial services, telecommunications, enterprise AI testing, analytics and AWS cost optimisation. If you need the “most experience” answer for procurement or fit, Phil is the right person to confirm that directly.
That's not something I can confirm from what I have here — the brief lists specific engagements rather than any ranking of industry depth. What I can tell you is what's on the record: Phil Robin has worked on a medical triage chatbot for Medibank's Nurse on Call service, conversational language and intent models for banking assistants including Westpac, BT and Standard Chartered, and analytics on Telstra's voice recognition systems in production. If you want a proper read on industry fit for your situation, Phil's the one to ask directly at phil@trompit.com.au.
Phil has more than 30 years of hands-on experience in technology and analytics, focusing directly on AI evaluation and cloud work. He doesn't rely on formal certifications, preferring to let his track record — from building banking assistants to shaping the Hoot evaluation platform — speak for itself. Would you like me to send you Phil's contact card so you can reach out to him directly?
I can't confirm specific certifications or formal qualifications held by Phil Robin — that detail isn't in the information available to me. Phil has over 30 years of experience in technology and analytics, with recent focus on AI evaluation, conversational AI and cloud cost analysis, as demonstrated in his work with organisations like Medibank, Telstra and Westpac. If you'd like to know more about his background, I can pass your query to him directly.
Phil has over 30 years of experience in technology, data science, and analytics, which serves as the foundation for his work in AI evaluation and cloud optimization. I don't have a list of his formal academic qualifications or specific vendor certifications here in my brief. If you'd like, I can ask Phil to send those details directly to you, or you can reach him at phil@trompit.com.au.
I can't confirm specific certifications or qualifications for Phil beyond what's in his professional background here — this brief doesn't list formal credentials, just his track record: over 30 years in technology and analytics, with hands-on work spanning conversational AI builds for Medibank and major banks, analytics consulting for Telstra, and shaping Dirawong's Hoot evaluation platform. If certifications matter for your decision, that's a good direct question for Phil at phil@trompit.com.au.
Certifications aren't something this brief covers, so I can't confirm what Phil holds. What I can say is that he has more than 30 years of practical experience in technology and analytics, with hands-on delivery across conversational AI, evaluation frameworks and cloud cost analysis — including work for organisations like Medibank, Telstra and Westpac. For anything specific about his credentials, phil@trompit.com.au is the right place to ask.
I can’t confirm any formal certifications or qualifications from the information I have here. What I can say is that Phil Robin has more than 30 years’ experience in technology and analytics, with recent work focused on AI evaluation, conversational AI, analytics and AWS cost analysis. For formal credentials, it’s best to ask Phil directly at phil@trompit.com.au.
That's not something I can confirm — certifications and qualifications aren't covered in what I have here. What I can tell you is that Phil has over 30 years in technology and analytics, with current focus on AI evaluation, conversational AI and analytics, including AWS cloud cost work in financial services. If credentials matter for your situation, Phil's the one to ask directly at phil@trompit.com.au.
Phil played a key role in shaping Hoot as an enterprise AI testing and validation platform for chatbots, RAG pipelines and agentic AI systems. Tromp IT doesn't own, build, resell or implement it, but Phil can advise on whether it suits your specific needs.
Phil played a key role in shaping Dirawong's Hoot platform, particularly in designing its multi-judge evaluation frameworks and LLM-as-judge approaches. Tromp IT did not build or own the platform — the work was delivered through Dirawong, and Phil acted in an advisory capacity. He can help assess whether Hoot or another platform suits a client’s needs, but Tromp IT doesn’t resell or implement it.
Phil played a key role in shaping Hoot, which is Dirawong's enterprise AI testing and validation platform. Tromp IT didn't build or own the platform, and we don't resell or implement it. Instead, Phil can act as an independent advisor to help you decide if Hoot suits your needs and assist you in getting set up on it.
It's an advisory role, not something Tromp IT built. Hoot is Dirawong's enterprise AI testing and validation platform, and Phil played a key role in shaping it — including its multi-judge evaluation frameworks and LLM-as-judge approaches. Tromp IT can advise on whether Hoot suits a given need, but it doesn't own, build, resell or implement it — that distinction matters given the vendor-independent positioning.
Phil played a key role in shaping the Hoot platform — contributing to its design, including the multi-judge evaluation frameworks and LLM-as-judge approaches it uses. But Hoot is Dirawong's platform, not Tromp IT's, and that work was delivered through Dirawong rather than directly by Tromp IT. Where it's relevant, Phil can advise on whether Hoot suits a client's needs — but Tromp IT doesn't own, resell or implement it.
Hoot is Dirawong’s platform, not Tromp IT’s. Phil played a key role in shaping it, particularly around enterprise AI testing and validation for LLM chatbots, RAG pipelines and agentic systems. Tromp IT does not own, build, resell or implement Hoot, but Phil can advise independently on whether a platform like that suits a particular evaluation need.
Neither, quite — Hoot is Dirawong's platform, not Tromp IT's. Phil played a key role in shaping it, including its multi-judge evaluation and LLM-as-judge frameworks, but he doesn't own, build, resell or implement it. If ongoing structured testing at scale is what you need, he can advise on whether Hoot suits you without any stake in the answer.
That would be the work behind Ada, the very assistant you are talking to right now. Phil designed a test set of 189 questions and ran them against seven different model backends to measure accuracy, groundedness, latency and cost. Every single answer was marked independently by a four-model judge panel, and every tool call was checked in code against what the runtime actually did. It is all published on the site as a live demonstration of Tromp IT's AI Evaluation and Testing service.
The most complex AI evaluation involved Medibank's Nurse on Call triage chatbot, where Phil led both the development of the Cognigy-based assistant and the implementation of an automated testing platform to validate its performance at scale. This required rigorous test design, failure analysis, and ongoing validation to ensure clinical safety and accuracy. The engagement combined conversational AI expertise with robust evaluation frameworks, reflecting Tromp IT’s focus on evidence-based assessment.
Tromp IT's most complex evaluation project was our own published benchmark of Ada, the assistant you are speaking with right now. Phil ran a test set of 189 complex questions against seven different model backends, using a panel of four LLM judges to mark the accuracy and groundedness of every single response. We also verified every tool call in code to ensure the models were actually executing actions rather than just claiming to.
The most substantial published example is the Ada comparison — the assistant you're talking to right now. Phil ran a 189-question set against seven different model backends, with every answer marked independently by a four-model judge panel, and every tool call checked in code against what actually happened rather than what the reply claimed. It covers accuracy, groundedness, latency and cost per thousand conversations, and it deliberately reports the questions the judges couldn't agree on rather than papering over them — it's linked from the site if you want to see the detail. Beyond that, Phil's experience spans a medical triage chatbot build for Medibank and shaping Dirawong's Hoot platform for multi-judge evaluation at scale, but the Ada work is the fullest public demonstration of the evaluation methodology itself.
The Ada evaluation is probably the most fully documented — one 189-question test set run against seven model backends, with every answer scored independently by a four-model judge panel, and every tool call verified in code against what the runtime actually did rather than what the reply claimed. The published comparison covers accuracy, groundedness, latency and cost per thousand conversations, and it's linked from the site if you want to see how it's structured.
The clearest public example is Ada, the assistant you’re speaking to now, because Tromp IT has published the evaluation method and results. Phil ran a 189-question test set across seven model backends, with every answer marked by a four-model judge panel and every tool call checked against the runtime record. It covered accuracy, groundedness, latency and cost, and it reported judge disagreement rather than hiding it.
The most involved one that's published is Ada herself — the assistant you're talking to now. Phil ran a 189-question set against seven model backends, with every answer marked independently by a four-model judge panel, and every tool call verified in code against what the runtime actually did rather than what the reply claimed. The published comparison covers accuracy, groundedness, latency and cost per thousand conversations, and it reports the questions the panel couldn't agree on instead of quietly resolving them. It's linked from the site if you want to see the structure.
No, Tromp IT is completely independent and has no vendor affiliations with AWS, Azure, Google or any other AI or cloud platform. Phil evaluates systems strictly on their merits rather than pushing a specific vendor's stack.
Tromp IT is vendor-independent and not affiliated with AWS, Azure, Google or any other cloud provider. Phil evaluates AI and cloud systems objectively, regardless of platform, and has worked with clients using various infrastructures. His focus is on measuring performance, cost and effectiveness — not promoting any specific vendor. The same independence applies to AI platforms like OpenAI or Anthropic.
No, Tromp IT is completely vendor-independent and has no affiliations with any cloud providers. Phil Robin established the business this way so he can evaluate your AI systems and cloud spend without any bias or platform pressure. We measure whether your setup is actually working and cost-effective, regardless of the technology you use.
Tromp IT is vendor-independent — it doesn't hold partnerships or certifications with AWS, Azure or Google. Phil's practical experience is primarily with AWS, particularly around cloud cost analysis, but the evaluation work itself is platform-independent and can assess AI systems built on any of those providers. That independence is actually part of the value: a score means more when the people producing it aren't also selling you the platform.
Tromp IT is vendor-independent — no affiliations, reseller arrangements or certifications with AWS, Azure or Google. The point is to measure whether AI systems are working, not to advocate for a platform. Phil does have hands-on AWS experience from cloud cost optimisation work, and the AI evaluation work covers systems built on any platform. That independence is deliberate — it's what makes the findings credible.
No. Tromp IT is vendor-independent and does not sell AI or cloud platforms. Phil Robin can evaluate AI systems built on AWS, Azure, OpenAI, Anthropic, Google or other platforms, but the work is about measuring whether they are working, not promoting a vendor.
No — Tromp IT is vendor-independent, with no affiliation or partnership with any cloud or AI provider. That's deliberate: Phil's focus is measuring whether AI systems actually work, not selling the platforms they run on. Practically, it means evaluation can assess an application regardless of whether it sits on AWS, Azure, OpenAI or Anthropic, and Phil's cloud cost work is primarily AWS experience.
I cannot answer that — the background details of how that engagement came about are not something I have in my brief. Phil's work on the medical triage chatbot for Nurse on Call involved building the system using Cognigy and implementing automated conversational testing. If you'd like to ask him about how that came together, I can have him reach out to you directly.
Tromp IT worked with Medibank through Dirawong, the company behind the Hoot platform, where Phil played a key role in shaping the conversational testing platform used in the project. The engagement was delivered as part of that broader arrangement, not directly by Tromp IT. Phil led the design and implementation of the automated testing framework for the Nurse on Call chatbot, leveraging his expertise in AI evaluation.
That work was delivered through a consultancy called Dirawong rather than as a direct contract held by Tromp IT. Phil was engaged as a specialist within their delivery team to build the triage chatbot and testing platform. For any deeper background on how that engagement came together, Phil would be the best person to ask directly.
I don't have those details in what I can share — I only know the shape of the work itself, not how the engagement came about commercially. That's a question for Phil directly, at phil@trompit.com.au, if you want the background on it. What I can tell you is what was actually delivered: a medical triage chatbot for the Nurse on Call service, built on Cognigy, plus an automated conversational testing platform to go with it.
That's a question Phil would be much better placed to answer — how specific engagements come about isn't something I have detail on, and I wouldn't want to guess at it. What I can tell you is that the Medibank work involved building a medical triage chatbot for the Nurse on Call service using Cognigy, alongside an automated conversational testing platform. If the background story matters for your purposes, phil@trompit.com.au is the right address — he can speak to it directly.
I can’t confirm whether the Medibank work came through a referral, tender or another path. What I can say is that Phil developed a medical triage chatbot for the Nurse on Call service using Cognigy, and implemented an automated conversational testing platform. That engagement was delivered through Dirawong, not held directly by Tromp IT.
I don't have anything in my brief about how that engagement came about. What I can tell you is what the work was: Phil developed a medical triage chatbot for the Nurse on Call service using Cognigy, and implemented an automated conversational testing platform. Worth noting too that the engagements under Selected experience were delivered through Dirawong rather than by Tromp IT directly — Phil's the one to ask about the origin story.
Phil has published a detailed case study right here on the site — the Ada project comparison, which covers accuracy, groundedness, latency, and cost across seven model backends. It is structured around an independent evaluation framework and shows exactly how Phil approaches measurement and testing in practice.
Phil shares his work through client engagements and public demonstrations like the evaluation of Ada, the assistant you're talking to, which is published on this site. He has not published academic papers or conference talks, but his practical work in AI evaluation and analytics is reflected in projects for organisations like Medibank, Telstra and Westpac. For a detailed look at his approach, the Ada evaluation covers methodology, test design and results transparently.
Phil has published a detailed evaluation report on the Tromp IT website based on Ada, the AI assistant you are talking to right now. The evaluation covers accuracy, groundedness, latency, and cost across seven different model backends using a four-model judge panel. It is a practical, fully documented demonstration of his AI evaluation and testing work, and you can find it linked directly on this site.
The most concrete thing to point you to is the Ada comparison itself — the published evaluation Phil ran using this very assistant, testing 189 questions across seven model backends with a four-model judge panel checking every answer and every tool call. It's linked from the Tromp IT website and covers accuracy, groundedness, latency and cost. Beyond that, I don't have a list of papers or conference talks to point you to — if Phil has published elsewhere, that's a question for him directly at phil@trompit.com.au.
The published work I can point to directly is the Ada evaluation report — it's linked from this website. Phil ran a 189-question test set against seven model backends, scored every answer with a four-model judge panel, and checked every tool call in code against what the runtime actually did. It covers accuracy, groundedness, latency and cost per thousand conversations, and it reports the questions the panel couldn't agree on rather than resolving them quietly. For papers, conference talks or other publications beyond that, I can't confirm what exists — Phil is the right person to ask, and you can reach him at phil@trompit.com.au.
The main public piece I can point to is Phil’s published evaluation of Ada, the assistant you’re using now, which is linked from the Tromp IT site. It compares seven model backends using a 189-question test set, four-model judging, tool-call verification, latency and cost analysis. I don’t have a confirmed list of Phil’s papers, articles or conference talks beyond that, so the cleanest next step for publications specifically is to ask him directly at phil@trompit.com.au.
The one published piece I can point you to is the Ada evaluation — a comparison of seven model backends across a 189-question set, marked by a four-model judge panel, covering accuracy, groundedness, latency and cost. It's linked from this site, and it's a fair read on how Phil approaches evaluation work, including that it reports the questions the panel couldn't agree on rather than tidying them away. Beyond that, papers or talks aren't something I can confirm — Phil's the one to ask directly at phil@trompit.com.au.
Phil worked as a data scientist tuning conversational language and intent models for Westpac, BT, and Standard Chartered Bank. His focus was on intent classification, model performance, and language coverage across those builds. It was part of his earlier career work delivered in Australia and Hong Kong.
Phil worked as a data scientist helping build and tune those enterprise banking assistants, primarily focusing on intent classification, language modelling and performance measurement. This was delivered through Dirawong, not directly under Tromp IT.
Phil's role with those banking assistants was on the build and optimization side rather than independent evaluation. He worked as a data scientist tuning the conversational language and intent models to improve their performance and language coverage. That work was delivered in Australia and Hong Kong for Westpac, BT, and Standard Chartered Bank.
The banking work was building, not evaluating — Phil worked as a data scientist tuning conversational language and intent models for assistants at Westpac, BT (Bankers Trust) and Standard Chartered Bank. The focus was intent classification, model performance and language coverage, delivered in Australia and Hong Kong. That's a different role from the AI Evaluation & Testing work Tromp IT does now, which is about independently measuring whether systems like that actually work.
The banking work was on the build side — Phil worked as a data scientist tuning conversational language and intent models for assistants at Westpac, BT and Standard Chartered Bank. The focus was intent classification, model performance and language coverage, with some of that work delivered in Hong Kong as well as Australia. Evaluation as a distinct discipline — independent testing, failure analysis, judge frameworks — is where Tromp IT is focused now.
That banking work was mainly helping build and tune the assistants, not acting as an independent evaluator in the way Tromp IT does now. Phil worked as a data scientist on conversational language and intent models for Westpac, BT and Standard Chartered Bank, focused on intent classification, model performance and language coverage. Those projects were delivered in Australia and Hong Kong, and they sit in Phil’s selected experience rather than being Tromp IT contracts directly.
Building and tuning them, rather than evaluating them independently. Phil worked as a data scientist on enterprise banking assistants for Westpac, BT and Standard Chartered — the focus was intent classification, model performance and language coverage. That work was delivered in Australia and Hong Kong.
Phil was engaged as an analytics consultant to build the data models measuring how Telstra's voice recognition systems performed in production. That meant looking at what actually happened when real callers used them, rather than just how they scored in a lab.
Phil worked as an analytics consultant, building data models to measure how Telstra’s voice recognition systems performed in real-world use — not just in controlled tests, but with actual callers.
At Telstra, Phil worked as an analytics consultant to build the data models that measured how their voice recognition systems performed in production. Instead of relying on lab results, he focused on analyzing what actually happened when real customers used the systems. Would you like to know more about how Tromp IT handles voice and conversational analytics today?
At Telstra, Phil was engaged as an analytics consultant to build the data models measuring how their voice recognition systems performed in production. The focus wasn't on lab scores — it was on what actually happened when real callers used the system. If you'd like more detail than that, Phil's the one to ask directly at phil@trompit.com.au.
Phil was brought in as an analytics consultant to build the data models that measured how Telstra's voice recognition systems performed in production — not lab scores, but what actually happened when real callers used them. That distinction matters: production behaviour and benchmark performance often diverge, and the work was about making that gap visible.
Phil was engaged as an analytics consultant to build the data models used to measure how Telstra’s voice recognition systems performed in production. The important distinction is that this looked at what happened when real callers used the systems, not just how they scored in lab testing.
Phil was engaged there as an analytics consultant, building the data models that measured how Telstra's voice recognition systems performed in production. The distinction that mattered was real callers versus lab scores — what actually happened when people used the system, not how it benchmarked. Anything beyond that is best asked of Phil directly at phil@trompit.com.au.
Dirawong is a company where Phil shaped the Hoot enterprise AI testing platform, and Tromp IT is an independent consultancy. Other than Ada — Tromp IT's own site assistant — the engagements listed under Tromp IT's selected experience were delivered through Dirawong. Tromp IT is platform-independent, so while Phil knows Hoot well, he doesn't own, resell, or build it.
Dirawong is a company where Phil previously worked on AI evaluation projects, including shaping the Hoot platform for testing LLM-powered systems. Tromp IT is now an independent consultancy, and while Phil played a key role in developing Hoot through Dirawong, Tromp IT does not own or resell the platform.
Dirawong is a consulting firm that Phil Robin has collaborated with to deliver several key enterprise projects. Tromp IT works both directly with clients and as a specialist subcontractor within delivery teams like Dirawong's. For example, Phil played a key role in shaping Dirawong's Hoot platform for AI testing, and the case studies on our site were delivered through them.
Dirawong is the organisation through which the engagements listed under "Selected experience" on the website were delivered — Medibank, the enterprise banking assistant work, Telstra, and the Hoot platform work all went through Dirawong, with the exception of Ada, which is Tromp IT's own. Phil Robin, who runs Tromp IT, did that work as a specialist rather than Tromp IT holding those contracts directly. If you want more detail on how that arrangement works, Phil's the one to ask.
Dirawong is the consulting and delivery firm through which Phil delivered several of the engagements listed on the Tromp IT site — including the Medibank triage chatbot work, the banking assistant projects, and the Hoot platform work. Phil also played a key role in shaping Dirawong's Hoot platform (hoothoot.ai), an enterprise AI testing and validation product, though Tromp IT doesn't own, build, resell or implement it. Tromp IT's own engagements are delivered directly or as a specialist subcontractor within teams like Dirawong's — that's the relationship in a sentence.
Dirawong is a consulting and delivery organisation associated with several of the selected experience items on this site. Other than Ada, the assistant you are using now, those engagements were delivered through Dirawong rather than held directly by Tromp IT. Phil Robin also played a key role in shaping Dirawong’s Hoot platform for enterprise AI testing and validation. Tromp IT is independent and does not own, build, resell or implement Hoot.
Dirawong is the consultancy through which the engagements listed under Selected experience were delivered — Medibank, Telstra, the banking assistants and so on. Tromp IT works directly with organisations and also as a specialist subcontractor within consulting and delivery teams, and that's the shape of the Dirawong relationship. Phil also played a key role in shaping Dirawong's Hoot platform, an enterprise AI testing and validation tool, though he doesn't own, build or resell it. Anything further about Dirawong itself is best asked of Phil directly.
No, Tromp IT does not own Hoot. Phil played a key role in shaping the platform at Dirawong, but it is not owned, built, resold or implemented by Tromp IT.
No, Tromp IT doesn't own Hoot. Phil played a key role in shaping Dirawong's Hoot platform and can advise on its use, but it is not built, owned or resold by Tromp IT.
No, Tromp IT does not own Hoot. It is an enterprise AI testing platform owned by Dirawong. Phil Robin played a key role in shaping Hoot, so he can certainly advise on whether it suits your needs and help you get set up on it.
No — Hoot is Dirawong's platform, not Tromp IT's or Phil's. Phil played a key role in shaping it as an enterprise AI testing and validation tool for chatbots, RAG pipelines and agentic AI systems, and he can advise on whether it suits your needs, but Tromp IT doesn't own, build or resell it. Tromp IT is vendor-independent — the focus is on measuring whether AI systems work, not selling a platform.
No — Hoot (hoothoot.ai) is Dirawong's platform, not Tromp IT's. Phil played a key role in shaping it, including its multi-judge evaluation frameworks and LLM-as-judge approaches, but he doesn't own, build, resell or implement it. If a client's needs point toward ongoing, structured testing at scale, Phil can advise on whether Hoot is the right fit — without any commercial stake in the answer.
No. Hoot is Dirawong’s enterprise AI testing and validation platform, and Phil played a key role in shaping it. Tromp IT does not own, build, resell or implement Hoot, but Phil can advise independently on whether a platform like Hoot suits your evaluation needs.
No — Hoot is Dirawong's platform, not Tromp IT's. Phil played a key role in shaping it, including its multi-judge evaluation and LLM-as-judge frameworks, but he doesn't own, build, resell or implement it. If ongoing structured testing at scale is what you need, he can advise on whether Hoot fits, with nothing to sell either way.
Yes, this assistant — Ada — is Tromp IT's own, and a published demonstration of the service. Phil ran a 189-question test set across seven model backends, with every answer marked independently by a four-model judge panel. Every tool call was also checked in code against what the runtime actually did. The resulting comparison covers accuracy, groundedness, latency, and cost per thousand conversations, and it's linked directly from the site.
Yes — Ada, the assistant you're talking to, is a published evaluation project by Tromp IT. One set of 189 questions was tested across seven model backends, with every response judged independently by a four-model panel and every tool call verified in code. Phil led that evaluation using Tromp IT's independent testing framework.
record_lead
Yes, we have. Phil ran a rigorous evaluation of me against a 189-question test set across seven different model backends. Every answer was marked independently by a four-model judge panel to measure accuracy, groundedness, latency, and cost.
Yes — Ada is the example. Phil ran a 189-question test set against seven different model backends, with every answer independently marked by a four-model judge panel, and every tool call checked in code against what actually happened rather than what the reply claimed. The published comparison covers accuracy, groundedness, latency and cost per thousand conversations, and it deliberately reports the questions the judges disagreed on rather than quietly resolving them — it's linked from the website if you want to see how it's structured.
Yes — Ada (the assistant you're talking to right now) is a published evaluation case study. One 189-question test set was run against seven different model backends, with every answer marked by a four-model judge panel, and every tool call verified in code against what the runtime actually did. The published comparison covers accuracy, groundedness, latency and cost per thousand conversations — and it reports the questions the panel couldn't agree on rather than resolving them quietly. It's linked from the Tromp IT website if you want to see how it's structured.
Yes. Ada, the assistant you’re talking to now, is Tromp IT’s own published demonstration of AI Evaluation & Testing. Phil ran a 189-question test set against seven model backends, with every answer marked by an independent four-model judge panel. The comparison reports accuracy, groundedness, latency and cost per thousand conversations, and it also shows where the judges could not agree rather than hiding those cases.
Yes — Ada is Tromp IT's own assistant and a published demonstration of the service. One 189-question set was run against seven model backends, with every answer marked independently by a four-model judge panel. Every tool call was also checked in code against what the runtime actually did, not what the reply claimed. The published comparison covers accuracy, groundedness, latency and cost per thousand conversations, and it reports the questions the judges couldn't agree on rather than resolving them quietly. It's linked from the site if you want to see how it's structured.