FA3STER

FA3STER was built for the Inter-IIT Tech Meet 13.0 Pathway High-Prep problem statement: a financial-document intelligence system where the answer is only useful if the reasoning path can be trusted.

Agentic RAG

I architected a multi-agent RAG pipeline that produced 92%+ accurate responses and rendered live workflow graphs while querying financial documents. The system separated retrieval, reasoning, metrics, business interpretation, and executive reporting into autonomous layers.

Evaluation

The agentic chunking setup reached 90.4% context precision, 84.9% recall, and 93.1% faithfulness. The application was containerized with Docker and built on Pathway’s dynamic ingestion pipeline for low-latency document updates.