Hey, I'm Mathieu
RSS FeedApplied AI & Automation Engineer at Digilac, a Swiss Atlassian Gold Partner. I build production systems that connect LLMs with business workflows and public data — from open-source MCP servers to multilingual AI platforms and Atlassian-to-ERP automation.
Alongside my role, I'm completing a work-study Diploma ES in Computer Science at CPNE and building Quoven, a source-backed product validation SaaS. I also run a multi-provider infrastructure lab and teach skiing.
More about me →
Featured
Quoven — Building the Evidence Layer for Product Decisions
9 min readA build-in-public look at Quoven, my SaaS that turns a freeform brief into a sourced, structured dossier on a product decision — the multi-probe research pipeline, the stack, and the decisions I'm proud of.
Recent Posts
One Week, Three Frontier Models — and a New Normal for AI Releases
9 min readFable 5 is back, OpenAI shipped GPT-5.6 Sol, Terra and Luna, and Grok 4.5 landed a day earlier — all in one week. A look at what each release actually is, and at the quieter pattern underneath: frontier models now launch through a government checkpoint.
Claude Fable 5 — Locked Out of a Model, Even Its Makers
6 min readI barely got to use Claude Fable 5 before a US export-control directive cut off access for every foreign national — including, in effect, the people who helped build it. A few honest thoughts on the potential I glimpsed, what happened, and why I hope it comes back the right way.
Claude Design — The First AI UI Tool That Doesn't Feel Like AI
8 min readA hands-on look at Claude Design: what it is, why Anthropic built it, the results that surprised me, the unmistakable feeling that a real designer was in the loop, and the two ways out — Figma to keep designing, or Claude Code to ship.
LLM Inference — Where Your VRAM Goes and How to Get It Back
10 min readA practical look at LLM inference: how context, KV cache, and model weights compete for VRAM, why Q8 on the KV cache is a free win, how quantization formats like GGUF and MLX compare, and why vLLM beats Ollama or LM Studio when you actually need throughput.