Optimization / Computer vision / Industrial AI
I'm Medhy Vinceslas, an independent senior data scientist. Through my company Myelink, I help industrial companies solve real-world problems: process optimization, operations research, factory-floor computer vision, generative AI. From scoping all the way to production.
Services
Process optimization, scheduling, industrial recipe recommendation: algorithms that decide better and faster, under your real-world constraints.
Scheduling and control agents for production and supply chain.
Quality control, rare-defect detection, workstation analysis, in real factory environments, with GDPR-native design.
Agentic LLM systems and retrieval over your domain data (LangChain / LangGraph), with rigorous evaluation and guardrails.
Measuring the true effect of a decision, beyond correlations: A/B testing, uplift modeling, counterfactual methods on observational data.
Personalization at scale: from matrix factorization to Graph Neural Networks, depending on what your data justifies.
PySpark / Databricks / Palantir Foundry pipelines, CI/CD, monitoring, FinOps: your models hold up in production, and cost less.
Case studies
Predictive models and optimization algorithms for recipe recommendation at Soufflet maltings, plus a FinOps overhaul of data ingestion pipelines.
RL agent for parts replenishment on production lines to reduce line stoppages, with supply-chain demand forecasting. Technical lead of a team of 3 to 5 data scientists.
Multimodal ML models and agentic LLM systems with retrieval over clinical and biomedical data (rare diseases), from scoping to deployment.
// Excerpts from salaried and freelance experience, detailed references on request.
Testimonials
"Thanks to his skills and expertise in data science, software development and MLOps, he was a key player in scaling the solution. His ability to understand problems, solve them and bring the team along makes Medhy a major asset to any data team."
"Even when projects were difficult, he made sure to do whatever it took to build a product that turned the client's vision into reality. He makes sure his work is clearly understood by clients, even non-technical ones."
"Medhy is an excellent problem solver with an impressive ability to work with complex data and use machine learning algorithms to deliver relevant results. I strongly recommend Medhy for any engagement requiring data analysis and machine learning skills."
"On top of his professionalism and reliability, he brings proven technical skills. As a fellow Data Scientist, I can only recommend his work."
References
Products & open source
Freelance mission matching platform: five platforms aggregated live, natural-language criteria, and relevance ranking against the resume with explanations. Designed, built and operated by Myelink, from first prototype to production.
Read the articleConversational search engine for local food: from a natural-language grocery list to nearby producers, through a LangGraph agent and PostGIS geospatial queries.
Read the articleOptimizer for company director pay: multi-objective constrained genetic algorithm over a full simulator of French taxation. A Pareto front, not a single imposed compromise.
Read the articleLocal-first agentic personal assistant: three-branch router, hybrid retrieval, calendars and banking through MCP, code-validated citations.
Read the article3D UAV simulation and autonomous drone control for path tracking, in Python.
Read the article
About
An engineer by training (Polytech Orleans), I have practiced applied data science for over ten years: computer vision deployed in Forvia plants and Reinforcement Learning for production scheduling, where I led a team of data scientists as technical lead, then process optimization for Soufflet maltings (InVivo), and AI applied to clinical trials at AstraZeneca. I founded Myelink in 2024 to carry this work as an independent.
Business workshops, use-case definition and success criteria.
Fast proof of value on your real data, within weeks.
Deployment, monitoring, documentation and testing.
Your teams grow throughout the engagement: code reviews, pairing, best practices.