profile = {
"name" : "Khalfa Youssef",
"education" : ["École Centrale de Lyon 🇫🇷", "École Centrale Casablanca 🇲🇦"],
"focus" : ["Machine Learning", "LLMs & RAG", "MLOps", "Optimization", "Data Engineering"],
"languages" : {"French": "C1", "English": "B2+", "German": "A1"},
"available" : "April 2026 — open to PFE internship (ML / Data Science)",
"contact" : "youssef.khalfa@etu.ec-lyon.fr",
}|
Built and trained a full GPT-style language model (char-level → BPE) using PyTorch. Val loss dropped from 7.79 → 5.07. Full pipeline: training, checkpointing, beam search, top-k/p sampling, temperature control. 
|
RAG + OCR + LLM platform that cut document retrieval time by 50% for ONCF maintenance teams. Stack: Tesseract, CLIP, Milvus vector DB, Llama 3.3. 
|
|
End-to-end predictive maintenance pipeline on multi-sensor time series. Windowing, labeling, grid-search over LSTM / Bi-LSTM / CNN vs. classical models for cobot stop prediction.
|
MILP formulation of the Split Delivery VRP + VNS metaheuristic. Solved small instances optimally in 0.6s, scalable to large instances via PuLP + custom heuristics. 
|
|
Compared KNN, MLP, and ResNet18 (transfer learning) for image classification, then shipped it as an interactive web app with FastAPI + React frontend. 
|
MATLAB/Simulink warehouse robot with LiDAR navigation + D* Lite path planning. Achieved −90% collisions and +35% reactivity vs baseline. 
|
| Period | Role | Company | Key Impact |
|---|---|---|---|
| 05–07/2025 | R&D Consultant — Industrial Methods Engineer | ESITH 🇲🇦 | +16% line efficiency via dynamic balancing; ML pipeline + Power BI dashboards; full-stack (FastAPI + React) |
| 07–08/2024 | Assistant Production Manager — Quality Control | SUPER CABLES 🇹🇳 | KPI dashboards with Power Query/VBA; ERP traceability (Sage X3); 3D station redesign (CATIA) |
Languages
ML / DL / AI
Data & Analytics
Web & Infra
Tools & Workflow
