>>> print("hello, world")
Data & AI engineer
I turn messy data into tools people actually use: pipelines, ML models, LLM assistants. Fewer slides, more results.
I build the plumbing between raw data and useful answers. And I make sure it doesn't leak.
Engineering student in Big Data & Machine Learning at EFREI Paris. At RiskHunter and YMUNIT I worked on data pipelines, RAG systems and LLM integrations for cybersecurity teams. I like to understand the models I use, down to the math when needed, and I favour solutions that are simple, measurable and explainable.
Off the keyboard
Volunteering
| field | value | |
|---|---|---|
| 0 | based_in | Nanterre, Île-de-France |
| 1 | school | EFREI Paris, Big Data & ML |
| 2 | looking_for | Work-study, 24 months, 3 days work / 2 days school |
| 3 | languages | French (native), English C2, Spanish B1 |
| 4 | certs | EBIOS Risk Manager, CS50P, Cisco Cybersecurity, PIX |
| 5 | in_progress | Microsoft AI-901, AI-300, AI-103, CS50 AI |
| 6 | award | Top scorer (40%), AXA Challenge 2025 |
EFREI Paris Panthéon-Assas, Villejuif
RiskHunter, Paris
YMUNIT, Paris
Les Sherpas, Nanterre
Atos, Dakar
Independent, Remote
CESI École d'Ingénieurs, Nanterre
Top of class 2025-26
Institution Sainte Jeanne d'Arc, Dakar
Highest honours
df: 30 projects
Security teams juggle 40+ compliance frameworks that all say the same thing in different words. This pipeline finds who says what: local embeddings throw out the obvious mismatches, an LLM (Mistral, Ollama) judges the rest. Result: 80% less prep time, and humans back on the interesting part.
Hospitals sit on mountains of data and still decide on gut feeling. This decision-support platform for a public hospital orchestrates the ETL with Airflow, refines the data through bronze, silver and gold layers, checks its quality and serves it in Metabase. Fully dockerized.
No TensorFlow, no PyTorch, just NumPy and backprop written by hand. A two-layer neural network that recognizes the digit you draw, live, in the browser. 84.4% accuracy, and above all the certainty of knowing what happens inside.
A file backup software built like a real commercial product: C# / .NET 8, WPF MVVM interface, versioned releases, GitFlow, documentation site and an external encryption module. School project, professional standards.
Messaging in real time, done properly: REST API, WebSocket, JWT authentication and JSON so messages arrive instantly and only to the right people.
F1 race analysis like a performance engineer, built on OpenF1.
Built during my work-study at RiskHunter: a Data/ML pipeline that ingests security data and has an LLM write MITRE ATT&CK / D3FEND attack scenarios, served through FastAPI with a Streamlit front. Thinking like an attacker, but automated. Code is private.
An online social deduction game, mobile-first: private rooms for 5 to 18 players, secret roles kept on an authoritative server, and a "constellation of suspicions" to place your doubts on the table. Built largely with AI coding agents, with tests on the game rules.
663
GitHub contributions, last 12 months
Each repo weighs the same, so one notebook can't drown the rest.
From first commit to last push. Open bars are still moving.
last sync: 14 Nov, 13:34 (api.github.com)
Kakoukaw is my other half: the one who looks at the light before looking at the numbers. Streets, skies, faces, no filter on the curiosity.
@kakoukaw
drag the film
I am looking for a 24-month Data, ML & AI work-study (M1 & M2, 3 days at work, 2 at school), available now. If your team has more data than time to use it, we should probably talk. Research collaborations and odd data problems welcome too.