January 19, 2026 · 9 min read
How to work with AI
A set of best pratices on how to work and collaborate with AI to co-own your work

Latest Articles
Thoughts on AI, Machine Learning, Data Science, and MLOps.
January 19, 2026 · 9 min read
A set of best pratices on how to work and collaborate with AI to co-own your work

January 18, 2026 · 4 min read
Mimesis is a web-based interview training platform. It allows master-level students and researchers to conduct realistic interviews with adaptive AI personas grounded in real interview data. The project combines two strong axes: rapid delivery using agentic AI coding practices, and deep conceptual work on persona design as a proxy for human social behavior.
Mimesis
August 1, 2025 · 5 min read
Un projet expérimental mêlant intelligence artificielle et humanités numériques pour traduire les œuvres de Molière en français moderne, vers par vers. Conçu comme un laboratoire de travail avec les LLM, il explore le prompt engineering, la génération contrôlée de texte et les enjeux d’accessibilité, à travers une application web et mobile pensée pour un usage pédagogique réel.
Moliere.love
August 1, 2024 · 3 min read
Building something that works is no longer the hard part. Knowing whether it should exist is. This piece looks at two AI-heavy projects that never shipped, not because the technology failed, but because business reality intervened. From toxic content detection on TikTok to a PMF validation platform for solo founders, both experiments used AI to compress time-to-insight, test assumptions early, and expose strategic constraints fast. The focus is on how AI reshapes early-stage decision-making by making it cheaper to be wrong — and clearer when it’s time to move on.

June 19, 2024 · 5 min read
Analyse multimodal des videos TikTok grace au deep learning et machine learning classique
August 1, 2023 · 3 min read
At Karhoo, I led pricing and forecasting for a global taxi marketplace operating in real time across multiple continents. Facing accuracy ceilings and scaling limits with bespoke scikit-learn random forest models, we benchmarked and migrated to Vertex AI AutoML on GKE, with feature pipelines in BigQuery and automated model promotion. The transition delivered significantly higher prediction performance, lower infrastructure and operational costs, built-in drift handling, and greater flexibility for global rollout under strict latency constraints.
