Projects – Nolan Cacheux
Selected among 7,000+ global applicants. Built an interactive political simulation powered by fine-tuned Mistral SLMs. 4 models fine-tuned via QLoRA in under 10 minutes each — our 8B outperforms Mistral Large on structured output at 10x lower latency. Production-level enterprise GenAI chatbot for Decathlon Finance, saving 13,000 hours/year for the suppliers team. Built with LangGraph, Google Gemini, and RAG architecture on GCP infrastructure. Production ML pipeline for predicting 8 sales KPIs across 64 sports categories using Prophet, PySpark, and Databricks with MLflow model registry. AI-powered autonomous quantitative analyst using LangGraph orchestration, hybrid RAG for SEC filings analysis, and real-time market data. Deployed on Azure Container Apps with Telegram bot interface. MLOps pipeline for detecting AI-generated product images using EfficientNet-B0. Features real-time FastAPI inference, MLflow experiment tracking, data drift monitoring, and Docker Compose deployment. Trained on the CIFAKE dataset (CIFAR-10 real vs Stable Diffusion AI-generated). ML-powered YouTube comment analysis tool with BERT sentiment detection, BERTopic clustering, and local LLM summaries via Ollama. Built with Next.js 15, FastAPI, and Transformers. Production-grade multi-tenant real-time multiplayer quiz platform built to handle 100+ concurrent users. Socket.IO WebSockets, JWT auth, Cloudflare CDN, and scalable Node.js backend on Hetzner VPS. Chrome extension integrating Stockfish chess engine for real-time position analysis, move evaluation, and intelligent suggestions. Built with TypeScript, Vite, and Chrome Extension Manifest V3. Autonomous AI-powered trading system using xAI's Grok model for market analysis and order execution via Interactive Brokers. Features real-time dashboard, automated trading loop, and AI self-reflection. Interactive 3D visualization platform for understanding Deep Learning architectures - from Perceptrons to Transformers. Built with Next.js 14, Three.js, and React Three Fiber. 
