Data Scientist & Computer Vision Engineer. Master's student in AI at EPITA (SCIA Major). I build end-to-end systems — not just models, but the pipelines, tooling, and infrastructure around them. From medical image segmentation to world models and LLM engineering, I care about what's happening inside the architecture, not just the final metric.
World-Models-Explorer Python PyTorch Gymnasium
World Models · Model-Based RL
From-scratch RSSM (Recurrent State Space Model) implementation. The model learns compact latent representations of environment dynamics and generates imagined future trajectories without interacting with the real world — core to Dreamer-style model-based RL.
MiniRAG FAISS sentence-transformers Ollama
Minimal RAG pipeline built from scratch — no LangChain, no abstraction layers. FAISS vector store, local embeddings, pluggable LLM backend. Built to understand what's actually happening at each step.
SentimentCam OpenCV HuggingFace MediaPipe
Real-time emotion analysis from live video. Face detection pipeline feeds into a transformer-based classifier, with emotion labels and confidence overlaid on the frame. CV meets NLP.
TextStyleTransfer PyTorch HuggingFace Gradio
Fine-tuned GPT-2 for formal ↔ informal style transfer. Includes full training pipeline, BLEU/BERTScore evaluation, and an interactive Gradio demo.
AI · ML · Vision
Web · Backend
Systems · Tools
