Pute

A real-time intelligence platform that transforms prediction markets from scattered prediction markets into actionable insights.

Comment

🎯 Pute - Prediction Market Intelligence Platform

💡 Inspiration

Prediction markets have exploded in popularity, with platforms like Polymarket seeing billions in trading volume. But there's a problem: how do you know which markets to trust?

Some markets are deep, liquid, and reliable. Others are thin, manipulated, or poorly designed. As traders and researchers, we constantly asked: "Is this market worth my attention?"

We built Pute to answer that question systematically. Instead of relying on gut feel, we wanted a data-driven way to evaluate market quality and uncover hidden opportunities.

🚀 What it does

Pute analyzes 100+ prediction markets in real-time and provides:

1️⃣ Trust Scores (0-100)

Every market gets a reliability score based on:

  • 💰 Liquidity depth
  • 📈 Trading activity
  • ⏰ Market maturity
  • 📉 Price stability
  • 📝 Resolution clarity
  • 💹 Bid/ask spreads

Output: Letter grades (A+ to D) so you instantly know which markets are trustworthy.

2️⃣ Opportunity Detection

Automatically identifies:

  • 🔥 Momentum Markets - Significant price moves backed by volume
  • 📊 Volume Spikes - Unusual activity indicating breaking news or whales
  • 💎 Value Plays - High-trust markets with extreme prices (potential mispricings)

3️⃣ AI-Powered Insights

Ask questions in plain English:

  • "What are the most trusted political markets?"
  • "Show me momentum opportunities"
  • "Find undervalued markets"

Get instant, data-driven answers.

🛠️ How we built it

Tech Stack:

  • Hex - Notebooks for data processing, app builder for dashboard
  • Python - Data pipeline and statistical analysis
  • Polymarket API - Real-time market data (100-150 active markets)
  • Statistical Methods - Percentile ranking, z-score normalization, sigmoid curves

Architecture:

Polymarket API → Data Processing → Statistical Scoring → 
Opportunity Detection → Interactive Dashboard + AI Chat

Key Components:

  1. Statistical Trust Engine

    • Built custom scoring algorithms using population statistics
    • No arbitrary thresholds - everything is relative to current market conditions
    • Used scipy for percentile ranking, z-scores, and normalization

  2. Opportunity Detector

    • Analyzes price movements, volume patterns, and trust scores
    • Filters false signals (price without volume = noise)
    • Ranks opportunities by statistical significance

  3. AI Query System

    • Pattern-matching NLP to understand user intent
    • Direct DataFrame querying for accurate responses
    • Formatted markdown outputs for readability

  4. Hex Dashboard

    • 4 interactive tabs: Markets, Opportunities, AI Chat, About
    • Real-time filtering and sorting
    • Clean visualizations with charts and tables

Development Time: ~20 hours (2.5 days)

😅 Challenges we ran into

1. Statistical Validity

Problem: How do you score markets without arbitrary thresholds?

Solution: Used population-relative statistics. Instead of saying "$1M liquidity = good," we use percentile ranking: "This market has more liquidity than 92% of all markets." Adapts automatically to market conditions.

2. Volume vs. Price Signals

Problem: Price movements alone are misleading - a market can move 20% on thin volume (noise) or high volume (real signal).

Challenge: We had to build volume-weighted confirmation to separate strong trends from false momentum.

Solution: Created a volume-weight metric and classified momentum as STRONG (high volume), MODERATE, or WEAK (low volume).

3. WebSocket vs. Snapshot

Problem: Should we stream real-time data or use snapshots?

Decision: Snapshots. For analytical insights (not trading execution), real-time streaming adds complexity without value. Judges care about insights, not millisecond updates.

4. Time Constraints

Problem: 20 hours to build everything - trust engine, opportunity detection, AI chat, and dashboard.

Solution: Prioritized ruthlessly. Built core engines first, then dashboard, then polish. Cut nice-to-haves (semantic models, websockets) to focus on what matters.

🏆 Accomplishments that we're proud of

✅ Novel Trust Scoring Algorithm
We didn't just add up metrics - we used proper statistical normalization (percentiles, z-scores, sigmoid curves). Mathematically rigorous and defensible.

✅ Volume-Confirmed Momentum Detection
Most tools just show price changes. We filter noise by requiring volume confirmation. This catches real trends early.

✅ Working AI Chat Without External APIs
Built intelligent query handling using pattern matching and DataFrame operations. Fast, reliable, no hallucinations.

✅ 100+ Markets Analyzed in Real-Time
Processing, scoring, and detecting opportunities across the entire Polymarket landscape.

✅ Clean, Professional Dashboard
Built a production-quality app in Hex with interactive filters, sortable tables, and clear visualizations.

✅ Finished on Time
Delivered a complete, working project with all core features in 20 hours.

📚 What we learned

Technical Learnings:

  1. Statistical methods matter - Percentile ranking and z-scores are far superior to arbitrary thresholds for scoring systems.

  2. Volume is crucial - Price movements without volume confirmation are mostly noise. Always validate signals.

  3. Hex is powerful - The combination of notebooks (backend) and app builder (frontend) makes rapid prototyping incredibly efficient.

  4. Simplicity wins - We almost over-engineered this with WebSockets and complex AI. The simpler snapshot + pattern-matching approach worked perfectly.

Domain Learnings:

  1. Prediction markets vary wildly - Political/economic markets average 82 trust score. Entertainment/celebrity markets average 58. Category matters.

  2. Liquidity is king - It's the strongest predictor of market reliability. Thin markets are easily manipulated.

  3. Trust ≠ Accuracy - A high trust score means the market is well-structured, not that the outcome is certain. Important distinction.

Process Learnings:

  1. Build core value first - We focused on the trust engine and opportunity detection before UI. Right call.

  2. Time-box features - Set hard limits. If a feature takes >1 hour and isn't core, cut it.

  3. Test incrementally - Building in cells let us validate each component before integrating. Saved hours of debugging.

🚀 What's next for Pute

Short-term (Next Month):

🔔 Alert System

  • Email/SMS notifications for high-priority opportunities
  • Custom threshold settings per user
  • Daily digest of top markets

📊 Historical Tracking

  • Track trust scores over time
  • Analyze which markets improve/degrade
  • Validate our scoring methodology against outcomes

📱 Mobile-Responsive Design

  • Optimize dashboard for phone/tablet
  • Quick-check interface for on-the-go analysis

Medium-term (3-6 Months):

🤖 Machine Learning Enhancement

  • Train models on historical market resolutions
  • Predict which markets will resolve accurately
  • Combine ML predictions with trust scores

🌐 Multi-Platform Support

  • Integrate Kalshi, Manifold, PredictIt
  • Cross-platform arbitrage detection
  • Unified trust scoring across platforms

⚡ Real-Time Updates

  • WebSocket integration for live data
  • Sub-second opportunity detection
  • Live price/volume charts

Long-term (6-12 Months):

🎨 Advanced Visualizations

  • Network graphs showing market correlations
  • Heatmaps of category performance
  • 3D trust-volume-price plots

🧠 AI-Powered Predictions

  • Use Groq with web search for outcome forecasting
  • Combine news analysis with market data
  • Generate probabilistic forecasts

💼 API for Developers

  • Public API for trust scores
  • Webhook integrations
  • Embeddable widgets for blogs/platforms

🏢 Enterprise Features

  • Custom market monitoring for platforms
  • Manipulation detection systems
  • White-label solutions for prediction market operators

🎯 Vision

We want Pute to become the Bloomberg Terminal of prediction markets.

Just as Bloomberg provides institutional-grade analysis for financial markets, Pute will be the trusted intelligence layer for prediction markets - helping traders make informed decisions and researchers understand market dynamics.

Prediction markets are the future of information aggregation. Pute makes them trustworthy.

Built with ❤️ for Hex-a-thon 2025

Built With

Share this project:

Updates

Submission history