OmniAlpha: Autonomous AI Quantitative Trading Desk

While most automated trading bots operate as opaque black boxes, OmniAlpha introduces an institutional grade quantitative trading desk built for total transparency, high velocity execution, and continuous AI learning. Developed for the hackathon using Alpaca Paper Trading, OmniAlpha runs a 2 second decision loop orchestrated by DeepSeek V4 NIM with instant Google Gemini Flash failover. To prevent execution halts during heavy market volatility or API throttling, we engineered a 3 chamber API key revolver that dynamically rotates credentials backed by a 4 second hard timeout. What sets OmniAlpha apart from conventional trading agents is our Live AI Prompt and Inference Inspector. Rather than hiding model decisions inside backend scripts, our terminal interface allows judges and traders to view the exact system prompts, real time market telemetry, and raw JSON inference outputs generated during every cycle. To eliminate repetitive trading mistakes, OmniAlpha implements Active Reflexion Memory. When a position closes, the engine evaluates the trade outcome and extracts a post mortem lesson. This lesson is stored in a persistent trade journal and injected directly into future LLM system prompts, allowing the AI to actively learn from past drawdowns. Combined with bi directional market analysis, dynamic ATR target exits, and strict 15 minute loss cooldowns that block re entry into falling assets, OmniAlpha delivers a resilient, audit ready quantitative platform built for the future of autonomous finance.