Agentic AI, built to survive production.
Book a working session →Who I work with
Developers
Who can code but are improvising their way through LLMs and agents, and want a structured mental model instead of a pile of prompts.
Teams
With a promising AI prototype that breaks, drifts, or burns budget the moment it meets real users.
Technical founders
Deciding what to build in-house vs. buy — and how to keep their data under their own control.
What we work on
Agentic architecture
Single-agent vs. orchestrated multi-agent, and when the extra complexity is actually worth it.
Tool design & function calling
Giving an agent the right capabilities without handing it the keys to everything.
Model strategy
Routing, fallback, local vs. hosted inference, and the cost/latency tradeoffs nobody puts in the demo.
Reliability & evaluation
Eval loops, guardrails, and observability, so you know it works before your users tell you it doesn't.
Retrieval
When RAG is enough, when you need agentic retrieval, and when you need neither.
Production integration
Fitting agents into an existing codebase and team, not a greenfield toy.
Sovereignty
Keeping sensitive data and inference on infrastructure you actually own.
Why me
Two decades of production engineering for clients who don't accept "it works on my machine" — Airbus, Cartier, SNCF, and Tate Britain (IK Prize 2016). I've built computer-vision and ML systems under those constraints, I work across the full stack, and I work in French and English interchangeably.
I'm not selling AI hype. I help you build the unglamorous parts — evaluation, guardrails, cost control, integration — that make AI actually hold up.
Formats
Quick call — 15 min
One question, one answer. Should I use X or Y? Is this the right problem to solve? Does this approach make sense? Good for a decision you can frame in two sentences.
Focused session — 30 min
A specific problem with a clear scope: reviewing a design doc, unblocking a stuck integration, choosing between two technical options. Come with context, leave with a direction.
Working session — 90 min
One concrete problem, solved or unblocked. Enough time to go deep: whiteboard an architecture, debug a live agent, or map out a roadmap you can actually ship.
Recurring mentorship
Ongoing relationship, your roadmap, async support between calls.
Team session
Architecture review or hands-on agent design with your engineers.
How it works
- 1
Pick a slot
Choose single session, recurring, or team session on the booking page.
- 2
Describe your problem in two lines
There's a field in the booking form. Be specific: what's broken, what you've tried, what you need to decide.
- 3
We get to work on the call
No sales pitch, no discovery round. 90 minutes on your actual problem.
- 4
If I'm not the right person
I'll say so and point you somewhere better.
Book a quick call
Pick a 15-minute slot below — no prep needed, just show up with your question.
Need more time? Book a longer session directly on cal.com.
Book a working session →