Alexandre
Girard Davila
I fix agentic AI that works in the demo and breaks in production.
Twenty years building systems that couldn't fail — Airbus, Cartier, SNCF, Tate. Now I do that for AI agents.
About
Senior Software Engineer with over two decades of experience in full-stack development, specializing in open-source solutions and AI/ML integration. Architected and deployed enterprise-scale platforms for computer vision, edge computing, and real-time data processing serving Fortune 500 clients.
Award-winning developer (IK Prize 2016 — Tate · Microsoft · Fabrica · Jolibrain) combining artistic vision with technical excellence in AI-powered installations. Active open-source contributor with widely-adopted tools for network visualization, computer vision, and hardware integration.
Download resume (PDF) →Mentorship
Agentic AI, built to survive production.
I help developers and engineering teams design agentic AI workflows that survive contact with real users. AI is heavily in the loop — but every session is a conversation with a person who has been on the hook for systems that can't fail.
// who I work with
Your agentic AI impressed everyone in the demo. Now it drifts, loops, leaks cost, or gives wrong answers the moment it meets real traffic. You don't need another framework tour. You need someone who has shipped systems that can't fail and can tell you fast whether you're looking at an architecture problem, an evaluation problem, or a model-strategy problem — and exactly what to do about it.
What we work on
The unglamorous parts that make AI actually hold up.
01
Agentic architecture
Single-agent vs. orchestrated multi-agent, and when the extra complexity is actually worth it.
02
Tool design & function calling
Giving an agent the right capabilities without handing it the keys to everything.
03
Model strategy
Routing, fallback, local vs. hosted inference, and the cost/latency tradeoffs nobody puts in the demo.
04
Reliability & evaluation
Eval loops, guardrails, and observability, so you know it works before your users tell you it doesn't.
05
Retrieval
When RAG is enough, when you need agentic retrieval, and when you need neither.
06
Production integration
Fitting agents into an existing codebase and team, not a greenfield toy.
07
Sovereignty
Keeping sensitive data and inference on infrastructure you actually own.
// why me
A human who has shipped the parts the demo skips.
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 don't sell the demo. I build the evaluation loops, guardrails, and cost control the demo skipped. You get a direct read on what's actually wrong and a path to fix it, from someone who has shipped this under real constraints.
Formats
One question, one answer — a real human on the line. Should I use X or Y? Is this the right problem? Frame it in two sentences and ring through.
A specific problem with clear scope: reviewing a design doc, unblocking a stuck integration, choosing between two options. Come with context, leave with a direction.
One concrete problem, solved or unblocked. Whiteboard an architecture, debug a live agent, or map a roadmap you can actually ship.
Ongoing relationship, your roadmap, async support between calls.
Architecture review or hands-on agent design with your team — not a greenfield toy.
How it works
01
Ring the hotline
Drop your question in the form. Two lines is enough — what's broken, what you need to decide.
02
I read it, not a bot
A real person with two decades of production experience — no triage queue, no chatbot deflection.
03
We talk it through
Fifteen minutes on your actual problem. No sales pitch, no discovery round.
04
If I'm not the one
I'll say so honestly and point you somewhere better. Your time isn't wasted either way.
The hotline
Let's have that coffee.
Think of it as a morning chat with an engineer who's already shipped this. One question, two lines, a real human voice on the other end.
No bot triage. No discovery call. If I'm not the right person, I'll say so and point you somewhere better.
Message sent.
I read every one personally — you'll hear back from a human.
Experience
Senior Software Engineer
Independent Consultant
July 2024 – Present
- Architected and deployed Colette, a production-ready self-hosted RAG and LLM serving platform.
- Streamlined deployment through containerization and UI optimization.
- Leading development of enterprise-grade open-source AI infrastructure.
Senior Software Engineer
Jolibrain
May 2018 – June 2024
- Architected React-based platform UI for the DeepDetect framework.
- Delivered edge-computing solution for Vinci, monitoring 50+ construction sites with 99.9% uptime.
- JoliGen AI integration for Cartier luxury retail.
- Built video metadata extraction system for railway footage and predictive maintenance.
Web Developer
Freelance
January 2011 – April 2018
- Won IK Prize 2016 at Tate Britain for 'Recognition' — comparing photojournalism with British art, with Fabrica and Microsoft.
- Engineered Raspberry Pi fleet management for Europa Group serving international conventions.
- Published open-source raspi-kiosk-server, adopted for digital signage.
Senior Ruby Developer
Legodata
January 2009 – January 2011
- Developed platform managing cultural events with Letsmotiv magazine.
- Designed scalable architecture for event data management and publication.
Projects
Colette
⭐ NewProduction-ready self-hosted RAG and LLM serving platform for enterprise AI infrastructure.
Visit Colette →Recognition
🏆 IK PrizeIK Prize 2016 winner at Tate Britain — AI-powered installation comparing photojournalism with British art. With Tate, Microsoft, Fabrica.
View on GitHub →JoliGen
⭐ FeaturedAI-powered generative art platform used by Cartier for luxury retail experiences.
Visit JoliGen →Parasol
⭐ 64A network graph exploration tool for visualizing and analyzing complex data relationships.
View on GitHub →React Bounding Box
⭐ 50HTML Canvas library to display bounding boxes on images with React, for computer vision.
View on GitHub →DeepDetect JS
⭐ 8JavaScript client library for the DeepDetect deep-learning server, for easy AI model integration.
View on GitHub →Get in touch
Prefer the slow lane? Drop me a line — or ring the hotline above for something quicker.
alex@girard-davila.net