Overview
Audit work is slow because auditors have to manually review every line item to decide what deserves attention. AuditGuard changes that flow. It uses machine learning to score and rank budget line items by risk, then uses an LLM to explain why each item was flagged — not just a score, but actual reasoning. The output is a prioritized list that auditors can act on immediately rather than starting from scratch. Built with FastAPI for the backend, Streamlit for the interface, Gemini API for explanations, and deployed via Docker on GCP Cloud Run.
Commissioned by a Masteral student at De La Salle University.
Gallery
