At Wild Vision, I built an AI-powered video post-production pipeline that combines Whisper transcription, YuNet face detection, and Demucs voice and music separation to turn long-form footage into social-ready cuts. I also developed a working prototype of an internal HR review tool with role-based access, a bilingual interface, and more than 100 automated tests.
As a freelance AI and Full-Stack Engineer, I designed and shipped a document Q&A RAG system using LangChain, GPT-4, and Pinecone, running at 500+ queries per day. I fine-tuned Mistral 7B and Llama 3 with LoRA, then added evaluation with RAGAS and BERTScore and Docker deployment.
My projects include a voice-controlled multi-agent plugin for Autodesk Revit and a RAG evaluation lab that improved a deliberately weak system from 6/12 to 12/12 through measured experiments. I’m focused on production LLM systems, including evaluation, monitoring, and cost and latency tracking.
