At the University of Utah, I built ORACLE, an open-source multi-objective reinforcement-learning optimizer that met 1,000 of 1,000 target specifications and reduced average optimization runtime from 85 minutes to 2.4 minutes.
I also engineered Agentic RAPTOR, a LangGraph-based autonomous design system that combines multimodal LLM reasoning, tool calling, planning, optimization, and SPICE verification. I developed MCP-based integrations with validation, retries, recovery paths, and human approval gates.
To evaluate and operate these systems, I built LLM and agent evaluation and observability pipelines, then deployed AI services using Python, FastAPI, Docker, Kubernetes, and AWS. I’ve also led technical discovery with customers and stakeholders, taking ambiguous workflows through prototyping, integration, validation, and deployment.
Earlier, at Dalian Maritime University, I developed Chinese-English Transformer pipelines for NLP tasks and optimized models with LoRA, PEFT, and quantization. At Uncle Fitter, I built customer-facing web and mobile applications and contributed to backend services with API response times below 200 milliseconds.
