Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
Inspiration
NGOs handle large volumes of financial documents, including invoices, receipts, purchase orders and project financial reports. Reviewing these records manually can be time-consuming, and inconsistencies or missing supporting evidence can be difficult to identify.
We wanted to build a smarter way to investigate financial records while preserving the integrity and traceability of the original documents.
What is VaultPilot AI?
VaultPilot AI is an intelligent financial evidence investigation platform designed for NGOs, finance officers, procurement teams and project managers.
Its core is a UGREEN NAS Skill that discovers authorized financial documents, extracts relevant information and reconciles records to identify inconsistencies, missing evidence and potential anomalies.
Every finding should be traceable to its supporting source documents, allowing a human reviewer to investigate and make the final decision.
How We Plan to Build It
The architecture is designed around four components:
- Frontend: A React dashboard for investigations, evidence tracking, findings and reports.
- Backend: Python and FastAPI for application logic and investigation orchestration.
- UGREEN NAS Skill: The core document discovery, extraction and reconciliation workflow using supported UGREEN capabilities.
- Storage: UGREEN NAS for authorized financial evidence, with PostgreSQL considered for application data and investigation history.
The initial implementation prioritizes the UGREEN NAS Skill and its evidence investigation workflow. The full-stack dashboard is part of the product development roadmap.
What Makes It Different?
VaultPilot AI focuses on evidence traceability rather than simply summarizing documents.
Its proposed Three-Way Evidence Ledger connects purchase orders, invoices and payment evidence, while financial report entries provide additional context. It classifies findings as confirmed discrepancies, potential anomalies, missing evidence or unverified items.
The system is designed to be read-only: original financial documents are not modified, and findings remain subject to human review.
Challenges and Learning
A major challenge is ensuring that financial findings are accurate, explainable and supported by actual documents. We are addressing this through conservative classification, source traceability, clear reporting of investigation limitations and careful use of the capabilities provided by the UGREEN Skill Developer Package.
Our goal is to demonstrate how NAS-native AI can make financial evidence investigations more organized, transparent and efficient.
What we learned
What's next for VaultPilot AI
Built With
- agentic
- ai
- api
- css
- fastapi
- nas
- postgresql
- python
- react
- rest
- sqlalchemy
- tailwind
- ugreen
- vite
