AI Engineering & Product
I build AI products in domains where a confident wrong answer has consequences. At I lead AI across the full stack, from retrieval and architecture to what ends up on screen. At I’m the technical co-founder, building a smart recovery diary for people dealing with long-lasting complaints.
of shipping things people actually use, first as a music producer, then as a frontend developer, now mostly in the space between a model and a product someone trusts.
Currently
Lefebvre Sdu
AI answers for legal professionals, drawn from an extensive legal database. I'm responsible for AI outcomes across the stack: retrieval quality, evaluation, latency, and interfaces that let a lawyer verify the answer instead of taking it on faith. Joined as a frontend developer in the AI team, now leading AI development across the organisation.
In production
Careblend
A smart recovery diary. One minute a day, so people with long-lasting complaints can see how the past weeks actually went instead of reconstructing it from memory in a ten minute consult. I handle the technology and development.
Closed validation with first practices
Position
Harness
Everyone can call the same endpoints. The difference is what you wrap around them: retrieval that surfaces the right passage, context assembly that survives real queries, evaluation that tells you whether last week's change helped, and a fallback that admits it doesn't know instead of inventing something. Swapping in a better model gives you a lift. Fixing the harness gives you a different product.
Provenance
In regulated professions the user stays accountable for the decision, so they need to be able to check the work. That's less about explainability as a research topic and more about plumbing: carrying provenance through the pipeline, being explicit about what wasn't found, and logging enough of the path that you can reconstruct an answer weeks later. The AI Act is pushing the whole market this way, but the people using these tools were already asking for it.
Evaluation
Model quality and product quality come apart faster than you'd expect. I've seen retrieval improvements nobody noticed, and interface changes with no model change behind them that shifted how much people trusted the thing. If you can't tell whether the output got better, you're not iterating. You're redecorating.
Archive
Five years of frontend development and product innovation consulting, mostly on discovery, prototyping and design systems. Different domain, same instinct: build the thing early enough to find out whether it's worth building.
Greenchoice EVApril 2023 - December 2023
A good example of how frontend development can be a catalyst for business transformation. By applying my technical skills to facilitate a data-driven, experimental approach, we not only achieved but exceeded our growth targets, setting a new standard for product innovation and market responsiveness.
PrivaOctober 2020 - Februari 2021
The Eco Building dashboard uses AI-driven forecasts based on a digital twin of the building to provide building managers with information about their building's performance.
NBD BiblionMarch 2023 - June 2023
The project's aim was to address the issue of low reading skills among youth in lower education schools, building upon previous qualitative research done with the target audience.
SduNovember 2019 - April 2020
The Titan Design System project at SDU was very important for my career, sharpening my skills and deepening my understanding of design systems in large-scale environments.
GreenchoiceOctober 2019 - July 2023
This project with Greenchoice was a pivotal point in my career, significantly enhancing both my technical and soft skills, and reinforcing my belief in the power of customer-centric development and agile methodologies for achieving business and user goals.
NBD BiblionNovember 2022 - Januari 2023
The Biblioscoop project with NBD Biblion was a challenge that not only showcased the feasibility of a library-focused streaming service but also demonstrated the critical role of customer feedback in shaping service offerings.
XpertHRJune 2021 - Februari 2024
A project with emphasis on balancing the need for ongoing product enhancement with the limitations of a legacy system, all while preparing for a significant technological transition.
What people said afterwards
“Thomas is your go-to guy when you need complex projects or product features made simple. His technical expertise, great communicative skills and organizational sensitivity helped me make the right choices as a Product Owner. Over the years I have worked with many developers, but Thomas is definitely one of a kind.”
“Thanks to Thomas, we have been able to test and investigate matters in a very efficient way. As a result, we avoid investing in a project at Greenchoice where the results ultimately did not align with the objectives.”
“Thanks to Thomas' expertise and approach, we have established a data-driven cycle of experiments in a short period of time. Linking customer needs to our goals is a very valuable way to identify opportunities for conversion improvement. We are now going to apply this method to other products and propositions.”
“We have been assisted in a very professional and efficient manner to gain insight into the demand for this new service among our target group. They have the knowledge and commitment to concretely help you further. Very involved and professional!”
