MarketLens: Enterprise Grade Financial RAG Engine MarketLens is a full-stack Retrieval-Augmented Generation (RAG) system designed to securely query complex financial documents (like Annual Reports).

Unlike standard AI wrappers, this project implements a custom, low level data ingestion and retrieval pipeline. It utilizes native Hugging Face models for privacy first, on device vector embeddings, and stores them in a local ChromaDB instance. The REST API is built with FastAPI, featuring strict multi-tenant architecture that safely injects user-provided Gemini API keys at runtime without compromising server state.

Key Engineering Highlights:

• Privacy First Embeddings: Uses SentenceTransformers (all-MiniLM-L6-v2) to run vector math completely locally, preventing sensitive document leaks to third party APIs.
• Manual Vector Orchestration: Bypasses high level framework abstractions to manually handle document chunking, metadata schemas, and distance based neighbor retrieval.
• Stateless Multi-Tenancy: The FastAPI backend securely handles per-request authentication, instantly garbage collecting LLM instances to prevent memory leaks and token cross contamination.
• Fully Typed & Documented: Utilizes Pydantic for strict schema validation and automatically generates OpenAPI (Swagger) documentation.