AutoHedge

Autonomous AI-Driven Quantitative Fund

A fully autonomous trading system that combines multi-agent LLM debate, neuroevolution-based portfolio allocation, and a hard-coded risk engine with absolute veto power. Every decision is auditable, every position is defined-risk, and no human intervention is required.

211
Python Files
36,515
Lines of Code
5
Uncorrelated Strategies
10
Risk Pipeline Checks

Three-Layer Architecture

Signals flow top-down through debate, allocation, and risk. Each layer can only restrict, never override, the layer below it.

Layer 3What to think about

Multi-Agent LLM Debate

Multiple Claude instances debate market theses with structured critique rounds. Consensus signals feed downstream only when conviction exceeds threshold.

3 debate agentsStructured critiqueConviction gating
Layer 2When to act

NEAT Meta-Allocator

Neuroevolution of Augmenting Topologies evolves neural network genomes to learn optimal strategy weights. No human-designed allocation rules.

Evolved genomesDynamic weightingNo human rules
Layer 1Absolute veto power

Risk Engine V2

10-check pipeline with hard limits on Greeks, correlation, drawdown, and concentration. Any single check can veto any trade, regardless of signal strength.

10 hard checksGreek limitsDrawdown veto
ExecutionVWAP / TWAP / Limit

IBKR Smart Router

All orders route through Interactive Brokers with algorithmic execution. Spread-only structures ensure defined risk on every position.

Smart routingSpread structuresDefined risk

Five Uncorrelated Strategies

Each strategy exploits a distinct market inefficiency. NEAT meta-allocation evolves the weighting across strategies to maximize risk-adjusted returns.

Topological Graph Arbitrage

Detects mispricing through mathematical structure of market relationships using persistent homology and graph diffusion.

Statistical Arbitrage

Cointegrated pairs with Ornstein-Uhlenbeck spread modeling and regime-break detection for mean reversion.

Cross-Sectional Momentum

12-1 month Jegadeesh-Titman factor with topological data analysis regime overlay for adaptive positioning.

NLP Sentiment

Claude API earnings analysis combined with NEAT-evolved signal weighting for information extraction alpha.

Volatility Risk Premium

Systematic VRP harvesting using spread structures only. No naked exposure by construction.

Competitive Edge

Structural advantages built into the system at the code level, not just operational policy.

No Naked Risk

SpreadType enum structurally excludes naked puts. Every option position is a defined-risk spread by construction, not by policy.

Evolved Allocation

NEAT genome learns optimal strategy weights through neuroevolution. Allocation is discovered, not designed by humans.

Regulatory Ready

EU AI Act compliant audit trail with SHA-256 hash chain. Every decision logged with full provenance for regulatory review.

Explore the Live Dashboard

View real-time portfolio metrics, trade history, regime analysis, and risk monitoring.

Enter Dashboard