Maximilien Rousseau

Hello — I'm a fourth-year engineering student at ESAIP, specializing in artificial intelligence.

Open to engineering and research internships in France or Europe, from April 2027.

I've always seen computer science as a tool, a skill worth having. A way to take a problem from a domain — like finance, industry, anywhere it matters — and turn it into something that works. That framing is what brought me to AI, for what it lets you build.

What I'm most worried about isn't a project failing, or AI taking over the world. It's letting my brain go rusty. Most of what's below is me trying not to.

Selected work

1. I.
   Sybella
   Eight weeks at a quantitative finance firm. Built a versioned market data pipeline that ingests 80 million rows in under an hour, after pivoting from Iceberg and Spark.
2. II.
   PatchCore
   Reproduced a 2022 industrial defect detection paper end-to-end. Reaches 99.03% accuracy on the standard benchmark, within 0.13 percentage points of the original authors' code.
3. III.
   MAS
   A ten-agent system that evaluates real-estate purchases using six public data sources. With 185 automated tests against the live APIs.
4. IV.
   Credit Scoring
   Full MLOps pipeline for credit default prediction. LightGBM tuned with Optuna, served via Docker, threshold optimized for actual business cost rather than accuracy.

See the full index of works.

Voir mon carnet PPP ING4.

How I work

Earn your complexity.

For each project, I start with an implementation plan, then break it down into workstreams and milestones. Each milestone becomes a set of task-level specs with the context, constraints, required skills, expected behavior, and acceptance criteria needed to implement it properly.

In practice, the workflow looks like this:
implementation plan → workstreams and milestones → repo-level instructions and skills → task creation → milestone implementation → tests, output review, and iteration.

I use LLMs and coding agents inside that structure. They help generate scaffolding, implement tasks, run tests, and move through milestones faster. My role is to keep the whole thing coherent: define the target, provide the right context, check whether the result matches the spec, and revise the task or the spec when something doesn't work.

What matters to me is understanding what was built, why it was built that way, how it was tested, and what should improve next.

What's next

First, a two-month summer internship in finance — a chance to spend more time on the data engineering side I started exploring at Sybella. Then six months on Erasmus at the University of Padua, in Italy, my second time studying abroad after Budapest last year. After that comes the part I care about most: a six-month engineering internship leading into my final-year project, the capstone of my five-year program.

Set in Newsreader and Inter. Built with Astro. Self-hosted at phitalys.fun.

▪