Looking for an internship · 2027

Many agents.
One decision.

M1 student in Distributed AI at Université Paris Cité. I build agents that search, plan and decide: game AIs, solvers and LLM assistants.

See my projects ↓GitHub ↗LinkedIn ↗

move your mouse · click to add an agent

01Selected projects

01 · 2026

Mr. Jack Pocket

The Mr. Jack Pocket board game, playable on computer, with an AI opponent at four difficulty levels.

My role: Team lead of a 6-person project. I wrote the entire AI and contributed to the game model and interface.

  • Minimax search with alpha-beta pruning, playing both sides (Jack and the investigator)
  • Evaluation function based on remaining suspects and estimated hourglasses
  • Benchmark harness: AI-vs-AI tournaments, pruning statistics, depth stability tests
4difficulty levels
19 msmedian time per move, depth 4
~3,700branches cut by α-β per move
JavaSwingSocketsGit flow
Code ↗
Mr. Jack Pocket game board
AI vs AI · 40 games per level
Easy
80%
20%
Medium
57.5%
42.5%
Hard
55%
45%
Expert
65%
35%
InvestigatorJack

02 · 2026

Jarvis

A personal assistant that manages my Google Calendar through a chat, powered by an LLM agent with tools.

My role: Solo project, ongoing.

  • Agent loop with tool calling: read, create, move and delete events, web search, long-term memory
  • Safety by design: risky actions need a confirmation card, enforced in code, not in the prompt
  • Imports a whole semester timetable (.ics) and creates events from a photo of a ticket or poster
2LLM engines (Gemini / local Ollama)
0destructive actions without approval
PythonFlaskGemini APIOllamaGoogle Calendar APISQLite
Code ↗
Move Thursday's lab to 2 pm
› reading calendar ✓

Update “AI lab” → Thu 2:00 pm?

approvereject

03 · 2025

Sokoban solver

A Sokoban game with animations and an automatic solver that finds the full solution by itself.

My role: Course project built on a teaching skeleton (G. Huard, GPL). The solver and AIs are mine.

  • A* search over game states, with simple deadlock detection to cut hopeless branches
  • My own data structures: priority queues (array and linked list), sequences, iterators
  • MVC architecture with the Observer pattern and animated moves
0.05 sto solve a Minicosmos level
JavaA*MVC
Repo private for now
Minicosmos 01

04 · 2026

Machine learning from scratch

Classic models re-implemented with NumPy only: no ML library for the learning part.

My role: Coursework, extended on my own.

  • SMS spam filter: logistic regression and Adaline trained with my own SGD
  • A neural network (MLP) with hand-written backpropagation
97.8%spam validation accuracy
95%MLP on two moons
92.9%MLP on handwritten digits
PythonNumPypandasJupyter
Code ↗
Decision boundary learned by the MLP

Also built

02About

I'm a first-year master's student in Distributed Artificial Intelligence (IAD) at Université Paris Cité. I like problems where a program has to choose: which move to play, which path to take, which tool to call. I enjoy taking an idea from algorithm to something people can actually use.

Education

  • 2026 – 2028Master's in Distributed AI (IAD)
    Université Paris Cité
  • 2023 – 2026Bachelor's in Computer Science
    Université Grenoble Alpes

Skills

Languages
PythonJavaCSQLJavaScript
AI / ML
Minimax α-βA* searchNumPyscikit-learnLLM agentsTool calling
Tools
GitFlaskSQLiteSwingJupyterLinux
Spoken
Turkish · nativeFrench · C1English · C1

03Let's talk

I'm looking for an internship in AI, multi-agent systems or software engineering. The fastest way to reach me:

LinkedIn ↗GitHub ↗