I build and optimize LLM systems. Active open-source contributor to the vLLM ecosystem (llm-compressor, compressed-tensors). I also develop production-grade LLM applications in RAG, fine-tuning, and autonomous agents.

vllm-project/llm-compressor : Quantization toolkit for LLM deployment with vLLM

• Add iMatrix weighted MSE observer and IMatrixGatherer : importance-weighted quantization, improves PPL across RTN/GPTQ/AWQ
• Add norm calibration context for unit-offset RMSNorm (Gemma/Qwen3Next) : fixes AWQ/SmoothQuant on Gemma models
• Add MoE calibration module for GlmMoeDsa (GLM-5) : packed 3D tensor handling for MoE architectures
• Fix topological ordering in FX graph cleanup : erase_node crash fix for Granite4 GPTQ
• Handle packed weights in granite4 to_3d_expert (W4A16)
• Fix SmoothQuant regex for DeepSeek/GLM-5
• Add SmoothQuant mapping for GLM-5
• Add AWQ mapping for GLM-5

vllm-project/compressed-tensors : Safetensors extension for sparse and quantized tensor storage

• Support N-dimensional tensors in pack/unpack_int32 : fixes 3D MoE expert weight packing

• finsight : Visual RAG for French financial documents using ColQwen2.5 + Qdrant + Claude Sonnet/Opus. Indexed 10 annual reports (~5,982 pages). 90% Recall@10, 100% citation accuracy. Async FastAPI backend with SSE streaming + background adversarial verification, React + base-ui frontend. 183 tests, CI/CD.

• reasonforge : Iterative LLM fine-tuning on Text-to-SQL using STaR (Self-Taught Reasoner). Ministral-8B: 60.1% baseline → 68.8% SFT → 78.0% after 3 STaR iterations on Spider dev set.

• ai-watch : Autonomous AI news agent via LangGraph + Claude tool-use. Aggregates HuggingFace Papers, GitHub Trending, and Simon Willison RSS into daily markdown briefings. Deployed on GitHub Pages, running via GitHub Actions.