AI-powered expense intelligence and automated fraud detection platform for small to medium-sized businesses.

Built by Team J2E2 (Jongmin Lee, Juan Felipe Duran, Elsa Zhang, and Édouard Chassé) in 24 hours for MPC Hacks 2026, a major hackathon organized by students from McGill University, Polytechnique Montréal, and Concordia University.

This project won 2nd Place for the Brim Financial Sponsor Track and Best Use of ElevenLabs from MLH, earning over $750 in prizes. Check out the original Devpost submission.

• Receipt OCR: Reads receipts directly from images using Gemini 2.5 Flash. Flags personal items, checks tip limits, and reconciles totals against card transactions.
• Fraud scoring: Evaluates transactions against a statistical baseline. Analyzes spending velocity, peer benchmarks, and historical patterns to assign risk scores and flag anomalies.
• Policy enforcement: Parses human-readable corporate spending policies into strict rules and enforces them against transactions in real-time.
• Voice assistant: An always-on assistant powered by Gemini 2.5 Flash, Whisper, and ElevenLabs TTS. Navigates the dashboard, queries spend data, and executes approvals via voice.
• Pre-approval workflows: Allows employees to submit budget proposals and routes them to managers with context and recommendations.

Python 3.12+ is required. A virtual environment is recommended.

1. Clone the repository:

   git clone https://github.com/DoudGeorges/cashflux.git
   cd cashflux

2. Install dependencies:

   uv pip install -r requirements.txt

3. Configure environment variables: Copy .env.example to .env and configure your API keys.

Start the Flask development server:

python app.py

Navigate to http://127.0.0.1:5000 and create a new account. To explore the pre-populated hackathon data, set your company name to Northwind Analytics.

Run the fraud detection unit tests:

python -m unittest tests.test_fraud_detector -v

MIT