DeltaMind_ai

DeltaMind AI is an autonomous options-trading agent built for the Alpaca AI Trading Agents Hackathon. It runs a real-time decision pipeline, orchestrated with LangGraph, that watches a curated basket of liquid, options-active symbols and evaluates one full decision cycle per symbol on every bar-close. The core design principle is gating cheaply before spending anything. A deterministic, hand-written technical screener has to confirm real momentum confluence, trend alignment on an hourly timeframe combined with a fresh crossover on a faster one, before a single LLM call is ever made. Only setups that clear this bar reach the AI layer, where Featherless AI (running Kimi K2 Instruct) acts as a lead strategist: not a rubber stamp, but a genuine contradiction check against live sentiment and news context. Every approved setup then passes through a second, entirely deterministic gate, a risk module with zero LLM calls in it, enforcing hard position-size, sector-concentration, margin, and stop-loss limits before anything reaches the brokerage. Nothing is ever forced through: a rejected setup is always logged with its exact reason, so the system's behavior is fully auditable, not a black box. Execution itself runs through Alpaca's official MCP Server, so every live order carries a structured, inspectable tool-call trail, while high-frequency market data uses direct REST for latency. A lightweight CLI wrapper handles low-overhead scheduled account checks. The whole system runs live and continuously against a fresh $100,000 Alpaca paper account, deployed on Render (backend) and Netlify (frontend), with a real-time dashboard for configuration, live logs, historical backtesting, and full decision-history transparency — including every rejected trade and the reasoning behind it.