FinDoc-Intelligence.AI is an enterprise-grade, asynchronous AI Agent platform built specifically for the [Google Cloud Rapid Agent Hackathon]
The platform solves a high-value corporate finance challenge: processing unstructured, high-volume financial documents (invoices, purchase orders, receipts) and turning them into validated, schema-compliant structures while executing autonomous compliance auditing and tracking financial leakages.
In modern corporate finance, verifying incoming financial documents against internal purchase orders, tracking tax compliance, and identifying duplicate billing is a massive manual bottleneck. Standard OCR solutions are brittle and fail to understand contextual line items.
This platform moves past simple linear pipelines to create a true Stateful Multi-Agent System that can:
- Ingest & Extract: Use multimodal
gemini-1.5-flashfor zero-OCR, high-speed structured schema extraction from raw document uploads. - Orchestrate: Leverage advanced graph-based workflows to route tasks across specialized agent nodes.
- Act & Verify: Interface directly with a live MongoDB Model Context Protocol (MCP) Server to read vendor profiles, track job states, and prevent financial anomalies before payment execution.
- Orchestration & AI: Google Cloud Vertex AI Agent Builder / LangGraph, Gemini 1.5 Flash & Pro
- Backend Framework: FastAPI (Python) with high-performance
asynciofor non-blocking task handling - Database & Integration: MongoDB (MCP Server Architecture) for resilient state persistence and transactional logging
- Environment Management:
uvfor lightning-fast, reproducible dependency management
To maintain a production-ready environment, this repository utilizes a strict two-tier branch system:
main: Reflects the production-ready, thoroughly tested MVP submitted for the hackathon.dev: The active integration branch where daily development, multi-agent updates, and experimental features are built and tested.