Building the next interfacebetween humansand intelligence.

Insight

The problem is not documentation. It is organizational memory. Teams should be able to understand what happened, why decisions were made, and how knowledge evolved over time, with the original evidence always within reach.

Approach

Transcibe turns conversations and documents into structured, connected organizational memory. It identifies decisions, commitments, people, risks, and reasoning, then connects them across meetings and systems so teams can ask questions and receive answers with timelines, context, and receipts.

Vision

Instead of repeatedly searching for information or reconstructing the past, organizations should be able to learn from it. Transcibe is building the memory and reasoning layer that allows knowledge to remain accessible, understandable, and alive.

Current Stage

Pre-launch, with the first pilot in development. I am building the product end to end, from knowledge architecture and AI systems to product design, infrastructure, and the way organizational memory should feel to use.

Insight

The gap between a working prototype and a production AI system is evaluation. Whoever measures, ships.

Approach

Full lifecycle ownership: data acquisition, model selection, development, validation with end users, deployment. RAG chatbot for the ANWB customer service center, live in production after iterative evaluation. Automated compliance checking for Dura Vermeer. OCR order processing for Onion Group. Generative content systems and team training for Shoeby. Intelligent client-location matching in care for BDO.

Lesson

Research, prototype, production, users: not a pipeline but a loop. Systems that skip the loop die quietly in staging.

Insight

If computers should adapt to humans, this is the hardest test: users who will never learn an interface. The interface had to be conversation itself.

Approach

Own initiative end to end: found the care facility myself, ran the interviews, built and deployed solo. Voice-driven chat that encourages conversation and suggests activities, sentiment and emotion analysis over time, and a dashboard giving caregivers early signals of declining well-being.

Lesson

Graded a 10. But the real result: watching someone in their eighties talk to software like it was a person. That moment set the direction for everything after.

Insight

Breadth first: frontend leadership at Pharmapartners, time-series forecasting at Informa, data engineering at ASML, sentiment analysis at Infoland, computer vision at Cre8ion.Lab.

Approach

Alongside: LLM fine-tuning, embeddings and vector databases, reinforcement learning experiments, sign-language translation with computer vision, MCP servers, AI agents. The lab notebook behind the production work.

Lesson

None of it was wasted. Production AI is 20% models and 80% everything I learned before models.