AI Financial Analyst

AI-powered financial intelligence that detects anomalies, uncovers risks, and transforms financial data into actionable audit insights.

Comment

Inspiration

Small businesses often spot money problems too late, like a duplicate invoice paid twice or a weekend wire nobody approved. Auditors are expensive and spreadsheets don't warn you. We wanted an affordable AI financial watchdog for any SME.

What it does

FinAuditor Enterprise reads your company's ledger and tells you how healthy your finances are.

  • Live dashboard: cash runway, monthly burn, gross margin, spending by category
  • Automatic red flags: duplicate invoices, weekend wires, SaaS price spikes over 50%, vendors with no W-9
  • Policy search so answers follow your own company rules
  • A chat assistant that knows your actual numbers
  • Executive audit reports with fix-it recommendations
  • Multi-company workspaces with CSV and policy uploads

Runway is estimated as:

\( \text{Runway (months)} = \frac{\text{Cash Balance}}{\text{Monthly Net Burn}} \)

How we built it

  • Backend: FastAPI, Python, SQLAlchemy, SQLite
  • Agents: LangGraph runs six agents in sequence
  • Policy search: ChromaDB with all-MiniLM-L6-v2 embeddings (local RAG)
  • Analysis: Pandas for ratios and runway projections
  • Frontend: HTML, CSS (dark glassmorphism), vanilla JS, Chart.js, Marked.js
  • Testing: Pytest
Planner → Retrieval → Analysis → Audit → Recommendation → Report

It runs fully offline with no API key, and OpenAI is an optional upgrade.

Challenges we ran into

  • Passing data between six agents without losing context
  • Keeping anomaly detection accurate, since too many false alarms get ignored
  • Keeping each company's data separate in a multi-tenant setup
  • Making AI answers stick to real ledger numbers instead of guessing
  • Running embeddings and vector search locally without slowing the app

Accomplishments that we're proud of

  • A full pipeline from raw CSV to executive audit memo
  • Rule-based and AI checks working together
  • A tool that works with no API key
  • A polished dashboard and sign-up portal
  • A tested backend, not just a demo

What we learned

  • Small, focused agents beat one giant prompt
  • RAG makes answers more trustworthy because they cite real policies
  • Simple rules catch most problems, and AI is best at explaining them
  • Good design and onboarding matter as much as the tech
  • Finance data needs careful handling from day one

What's next for FinAuditor Enterprise

  • QuickBooks, Xero, and bank feed integrations
  • Email and Slack alerts
  • Anomaly detection that learns each company's normal patterns
  • Role-based access for CFOs, auditors, and staff
  • Forecasting and what-if scenarios
  • PDF audit exports and a mobile app
Share this project:

Updates