.png&w=256&q=75)
🤓 Latest Submissions
Aegis Alpha
Inspiration In financial markets, AI "hallucinations" are financially fatal. We wanted to leverage the incredible reasoning power of modern LLMs, but put them in a strict mathematical straitjacket to preserve capital. Aegis Alpha bridges generative AI and quantitative finance, creating an autonomous agentic system you can actually trust to trade complex options. What it does Aegis Alpha is an institutional-grade, fully autonomous algorithmic options trading engine. It evaluates, structures, and executes multi-leg strategies (like Iron Condors) via the Alpaca API. Operators monitor the AI's internal thoughts, live Greeks, and Value at Risk (VaR) through a real-time telemetry dashboard. How we built it We designed a Tri-Agent Architecture to prevent rogue decisions: Intel Agent: Queries Alpaca for live options chains and financial news, performing real-time sentiment analysis. Debate Protocol: To eliminate hallucinations, we pit multiple LLMs (e.g., Groq and Llama-3.1) against each other. One proposes a strategy, while a challenger scrutinizes it. A trade only advances upon strict consensus. Risk Gatekeeper: The engine computes the Probability of Profit (PoP) using Black-Scholes pricing. If the PoP is below 55%, the trade is vetoed—overriding the AI. Position sizing is dynamically scaled using the Kelly Criterion and live VIX. The Executor then selects optimal strikes and legs into the trade. Challenges we ran into To avoid unhedged exposure, we deeply integrated with Alpaca’s MLEG (Multi-Leg) order class to ensure complex spreads filled as a single atomic unit. We also built a robust state reconciliation engine, parsing complex OCC option strings back to base tickers to keep our database perfectly synced with the broker, preventing wash-trade violations. What's next Expanding to dynamic Delta-neutral hedging and Iron Butterflies.
4 Sep 2026
