workknowgraph
Bioinformatics KG extraction
Multi-agent system for normalizing experimental assays
Research with Dr. Tingting Yu and Dr. Jinze Liu (UConn). The question: can we extract the process of experiments — the actual sequence of steps, samples, assays — across thousands of bioinformatics papers, and put them into a shared graph?
I orchestrated LLMs to normalize assay types via a custom multi-agent architecture: an Ontology agent grounds vocabulary, a Top-K agent retrieves candidates, a RAG agent contextualizes, and an LLM-as-a-Judge agent decides. Visualized with React + Vite.
Highlights
- Paper in progress on agentic systems for experimental-process extraction.
- Visualized knowledge graphs over 1000s of papers (React + Vite).
- Custom multi-agent: Ontology + Top-K + RAG + Judge.