Freelance · March 2025 – June 2025
Agentic AI Financial Intelligence Platform
A RAG pipeline for financial document intelligence — parsing uploaded documents and answering queries with source-grounded insights.
Project Overview
Built a document ingestion and retrieval pipeline: PDF parsing (PyPDF2), text chunking, and embedding generation via SentenceTransformer (all-MiniLM-L6-v2), indexed in FAISS for top-k retrieval. Retrieved chunks are used to construct grounded prompts passed to the Gemini API for response generation, backed by a FastAPI service layer for transaction management.
Key Features
- •Document ingestion: PDF parsing (PyPDF2), text chunking, embedding generation
- •FAISS-indexed top-k retrieval with grounded prompt construction for the Gemini API
- •FastAPI backend for transaction management with automated Debit/Credit consistency validation
- •React dashboard for uploading documents, submitting queries, and viewing AI-generated reports
Technology Stack
- •Python, FastAPI, React
- •PyPDF2, SentenceTransformers (all-MiniLM-L6-v2)
- •FAISS, PostgreSQL
- •Gemini API, JWT