Truth Serum: Potion Fraud Detector

Catching dishonest witches, one potion at a time

Comment

Inspiration

We were inspired by the real-world problem of fraud in supply chains and logistics. When we saw the HackUTD challenge about a potion factory with dishonest witches, we realized this was the perfect opportunity to build a fraud detection system that accounts for physics! The magical theme made it fun, but the underlying math and data science are serious — this same algorithm could detect fraud in oil pipelines, water distribution, or any system with continuous flow during collection.

What it does

Truth Serum is a fraud detection dashboard that catches dishonest courier witches who misreport potion amounts. It:

  • Analyzes minute-by-minute cauldron level data from the HackUTD API
  • Calculates unique fill rates for each of 12 cauldrons using time-series analysis
  • Detects drainage events and matches them to transport tickets
  • Accounts for continuous potion inflow during collection (the key insight!)
  • Flags suspicious tickets with 10–25% discrepancies and fraudulent tickets with >25% errors
  • Tracks witch trust scores (starting at 100, losing points for fraud)
  • Visualizes everything in a beautiful React dashboard with interactive maps

How we built it

Backend (Python/Flask)

  • Built a DataProcessor class that calculates per-cauldron fill rates from historical data using NumPy
  • Created a FraudDetector class that finds daily drain events (peak-to-valley analysis)
  • Implemented the critical formula:
    Expected = Visible Drain + (Fill Rate × Drain Duration)
  • Handled edge cases like multiple witches visiting the same cauldron per day
  • Served everything through a Flask API with endpoints for tickets, witches, and analysis

Frontend (React)

  • Created 4 tabs: Overview, Tickets, Witches, and Factory Map
  • Used Recharts for data visualization and Leaflet for an interactive factory map
  • Implemented real-time data fetching with auto-refresh capability
  • Designed a purple factory-themed UI with smooth animations and clean layout

Algorithm Design

  • Debugged and fixed the “Expected > 100” issue by dividing total drainage by the number of daily tickets
  • Tuned validation thresholds (10% / 25%) and trust penalties (-2 / -8)
  • Verified that all data comes from live API calls, not hardcoded assumptions

Challenges we ran into

  • The "Expected > 100" Problem: Our algorithm initially produced unrealistic expected amounts. We discovered overlapping witch visits and corrected it by adjusting for ticket counts.
  • Gradual vs Sudden Drains: We had to switch from discrete drain detection to daily peak-to-valley analysis.
  • All Witches at 0 Trust: Early penalties were too harsh; even honest witches lost all trust. Rebalanced to -2/-8.
  • Understanding Continuous Flow: Accounting for potion inflow during drainage was tricky but essential.
  • First-Time Coder: Learning React, Flask, and deployment simultaneously was a massive but rewarding challenge!

Accomplishments that we're proud of

✅ Mathematically sound algorithm that models real-world physics (continuous flow)
✅ Realistic fraud detection — 61% valid, 17% suspicious, 21% fraudulent
✅ Trust scores that evolve logically based on behavior
✅ Visually appealing UI with interactive maps and charts
✅ Modular design — data processing, fraud detection, and API layers are cleanly separated
✅ Robust handling of edge cases and real-time API integration
✅ Built a complex full-stack system

What we learned

Technical Skills

  • Designing algorithms that account for physical systems
  • Performing time-series analysis and anomaly detection in Python
  • Structuring RESTful APIs and handling real-time data
  • Building dynamic React dashboards with state management
  • Using visual cues (color coding, charts) for clarity in data representation
  • Balancing detection thresholds for accurate results

Problem-Solving

  • Sometimes “bugs” are real-world insights (expected > 100)
  • Learned iterative debugging with detailed logging and test scripts
  • Validated models with real data, not assumptions

Soft Skills

  • Explained complex math simply for judges and teammates
  • Documented everything clearly (README + QUICKSTART)
  • Learned to reverse-engineer from challenge specs to algorithmic design

What's next for Truth Serum

Bonus Challenge — Route Optimization

  • Find the minimum number of witches needed to prevent cauldron overflow
  • Create a scheduling algorithm using travel-time data
  • Factor in 15-minute unload times at the market
  • Visualize optimized routes on the map

Enhanced Features

  • Historical playback: visualize potion levels over time
  • Predictive analytics: forecast potential overflows
  • Fraud clustering: detect behavioral patterns among witches
  • Exportable PDF fraud reports
  • Sound alerts for fraud events
  • Dark mode toggle

Real-World Applications

  • Apply to oil pipeline or water distribution fraud detection
  • Use for warehouse inventory or shipping verification systems
  • Any domain involving continuous flow + periodic collection!

Built With

Share this project:

Updates