OmenAI (pronounced "omen eye")

🧠 Inspiration

Prediction markets are only as valuable as they are fair. Platforms like Polymarket allow anyone to bet real money on real-world outcomes - FDA approvals, mergers, elections, and even military events.

But what happens when someone already knows the answer?

If insider trading is illegal on Wall Street, why should it be any different when the same information asymmetry exists on-chain?

The key difference is transparency. Blockchain gives us a complete, immutable record of every transaction.

OmenAI turns that transparency into accountability.

🚀 What It Does

OmenAI is a real-time surveillance dashboard that detects potential insider trading on prediction markets.

It:

Ingests live trade data from Polymarket
Scores every wallet across multiple fraud signals
Uses machine learning + AI to detect anomalies
Generates plain-English explanations for flagged wallets

Think: SEC-grade surveillance tooling, but for on-chain prediction markets.

⚙️ How We Built It

📊 Data Pipeline

Pulled market + trade data from Polymarket Gamma + Data APIs
Stored trades in PostgreSQL
Deduplicated using transaction hashes

Risk Tier Classification

Tier	Category	Multiplier
1	FDA, mergers, government, military	1.5x
2	Sports, legal, macro	1.2x
3	Crypto, entertainment	1.0x

🔍 Three-Phase Fraud Detection

Phase 1 — Instant Scoring

Each wallet is scored based on account age, win rate percentile, avg bet size, topic concentration and speed-to-trade.

Phase 2 — Enrichment

Top 200 suspicious wallets selected
Full trade histories fetched using blockchain data
Cached in Redis (24-hour TTL)

Phase 3 — ML + AI Scoring

We apply Isolation Forest for anomaly detection and assign risk tiers:

$$ \text{Risk Tier} = \begin{cases} \text{High} & S \geq 80 \\ \text{Medium} & 30 \leq S < 80 \\ \text{Low} & S < 30 \end{cases} $$

AI generates human-readable explanations for flagged wallets

🖥️ Frontend

Built with Next.js 15 + React 19

Features:

Live SSE-streamed trade feed
Radar visualizations
Wallet deep-dives
Real-time fraud analysis runner with progress tracking

⚠️ Challenges We Ran Into

API Fragmentation
No unified wallet endpoint → combined multiple APIs with rate-limit handling
No Ground Truth
No labeled insider trading dataset → relied on signal design + unsupervised ML
Non-Blocking Pipelines
Needed real-time streaming + async processing → built SSE-based architecture

🏆 Accomplishments We're Proud Of

Built a real-time fraud detection pipeline end-to-end
Combined rule-based + ML detection effectively
Delivered explainable AI outputs, not just scores
Created a polished live dashboard under hackathon constraints

📚 What We Learned

Fraud detection is probabilistic, not binary
The hardest part is signal calibration, not model building
Combining heuristics + ML significantly improves robustness
Blockchain transparency is an underutilized regulatory advantage

🔮 What's Next for OmenAI

Wallet Graph Analysis
Detect coordinated trading rings via fund flow tracking
Real-Time Alerts
Email/webhook alerts for high-risk trades
Backtesting Engine
Validate against historical events with suspected insider activity
Cross-Market Expansion
Extend to Kalshi, Manifold, and more
Regulatory Reporting
Export structured compliance-ready reports
Community Flagging
Combine crowdsourced intelligence with ML detection