Agentic AIRVITMMay 2025 to Dec 2025
Sanjeevani, multi-agent RAG for Indian medicinal plants
Questions about medicinal plants span pharmacology, where a plant grows and whether it is endangered. One retriever can't answer all three, so a router sends each question to a specialist agent with its own index.
My roleFinal-year project, team of four; I built the agent system
The router sends 42 of 45 test questions to the right agent, and retrieval answers in a median 0.12 s.
The problem
Knowledge about Indian medicinal plants splits three ways: what a plant does, where it grows and how threatened it is. A single retriever mixes them up.
What I built
A LangGraph graph with a planner, a router and three specialist agents (research, GIS and IUCN conservation status), each backed by its own Weaviate collection, with a query-rewriting loop and a synthesiser. Llama 3.3 70B plans, Llama 3.1 8B routes and synthesises, and Whisper turns voice questions into text. The Streamlit interface shows answers with district maps and plant images.
Results
On the evaluation set the router sent 42 of 45 single-domain questions to the right agent (research 15/15, GIS 13/15, IUCN 14/15). Retrieval took a median of 0.12 s per query, before language-model time.
Related paper
Co-authored survey: “Agentic AI Approaches for Indian Medicinal Plant Knowledge Systems”, 9th International Conference on Computational System and Information Technology for Sustainable Solutions (CSITSS 2025), IEEE.
- LangGraph
- Weaviate
- Llama 3.3
- Whisper
- Streamlit