At Runware, I engineered diffusion-based FastAPI APIs that sustained 300,000+ production image and video generations across concurrent GPU models. I reduced inference latency by implementing parallel multi-GPU pipelines and model-layer caching.
At Vision RD, I designed a multi-head model for automated PCB inspection that achieved 98.2% defect-detection accuracy across four defect classes. I also built production RAG and function-calling assistants grounded in company content and live databases, using tools including FastAPI, Gemini, and Pinecone.
