BdsLab Multi-Agent Assistant
Bachelor's Thesis · University of Camerino
- Built a conversational multi-agent assistant that lets researchers drive six heterogeneous RNA-analysis command-line tools through plain natural language, removing the need to learn each tool's formats and options.
- Designed a supervisor-worker architecture in LangGraph that interprets a request, routes it to the right specialized agent, and chains tools into multi-step pipelines when a task depends on previous output.
- Implemented a RAG layer over a documentation knowledge base (ChromaDB + embeddings) to answer questions about the tools and the PhyloRNA taxonomy database.
- Made the system declaratively extensible (a new tool takes four localized steps, no changes to the graph or shared state), packaged it with Docker, and wrote a pytest suite with LLM calls mocked for deterministic tests.
- Python
- LangChain
- LangGraph
- ChromaDB
- Streamlit
- Pydantic
- pytest
- Docker
