FinsTrack

An AI-powered B2B operational expense engine that transforms physical receipts into a normalized USD cash flow ledger with human-in-the-loop verification.

Comment

Inspiration

Tracking physical receipts in operations like manufacturing or construction is a headache. Field managers spend hours manually typing paper invoices into spreadsheets, while traditional OCR tools constantly fail on faded ink or unusual receipt layouts. We built FinsTrack to solve this—turning messy paper receipts into a clean, automated operational expense log.

What it does

FinsTrack makes logging business expenses fast and painless:

  1. Upload: Drag & drop paper receipts directly into the web dashboard.
  2. AI Extraction: Gemini 2.5 Flash reads the receipt, extracting merchant, date, amount, and category into standardized USD.
  3. Human Review: Audit and edit the extracted details side-by-side with the original image.
  4. Live Analytics: Confirmed transactions instantly sync to our database, updating cash log tables and real-time category charts.

How we built it

  • Frontend: Built with Vanilla JS, HTML5, and Tailwind CSS for speed and zero build setup.
  • AI Engine: Google Gemini 2.5 Flash API configured with structured JSON outputs.
  • Backend: Supabase (PostgreSQL for transaction logs + Cloud Storage for receipt files).
  • Analytics: Chart.js for client-side cash flow visual breakdowns.

Challenges we ran into

  • Prompting Vision AI to handle crumpled, low-light paper receipts reliably.
  • Keeping the UI fast and smooth while handling cloud uploads and AI extraction simultaneously.

Accomplishments that we're proud of

  • Built a fully working B2B dashboard without heavy frontend frameworks.
  • Achieved a seamless end-to-end pipeline: Upload → Vision AI → Human Audit → Database → Real-Time Analytics.
  • Saved 90% of manual entry time while keeping a human-in-the-loop to ensure 100% data accuracy.

What we learned

  • How to effectively constrain multimodal LLMs for structured document processing.
  • Combining core SQL schemas with flexible PostgreSQL JSONB columns for dynamic receipt metadata.
  • Understanding Human Audit is still important and didn't make 100% decision for AI Model

What's next for FinsTrack

  • One-click CSV/Excel export for accounting integration.
  • Automated multi-currency conversion into USD.
  • Multi-user permissions for team approvals.

Built With

Share this project:

Updates

Submission history