
ADQuant is a production-grade autonomous agentic options trading desk that identifies, evaluates, and executes US equity options trades with zero human intervention. ARCHITECTURE A 10-layer LangGraph pipeline runs continuously: market ingestion → feature computation → 27 strategy micro-agents → confluence tournament → DeepSeek reasoning agent reasoning → 5-gate risk matrix → Alpaca MCP execution → 3-level exit guardian → email reporting. ALPHA GENERATION Four core quantitative strategies scan 500+ assets across 1D, 4H, 2H, and 30-minute timeframes: Liquidity Sweep Absorption, Trend Pullback Continuation, Lead-Lag Cross-Asset Propagation, and RSI Oversold Momentum Reversal. Signals that fire simultaneously across multiple independent strategies receive confluence bonuses (+25 for triple, +15 for double), then reasnoning agent writes an institutional investment thesis for the top setups and selects the best 3–5 trades. OPTIONS ENGINE Black-Scholes Greeks engine targets Delta 0.60–0.75 with 21–45 DTE. IV-regime routing selects optimal structure: long calls/puts (IV rank < 35), bull/bear spreads (35–55), short puts or bear call spreads (IV rank > 55). RISK MANAGEMENT 5-gate entry filter: (1) LLM confidence ≥ 75%, (2) IV rank regime check, (3) DTE window 21–45 days, (4) liquid US equities only, (5) Quarter Kelly dynamic sizing capped at 3% equity per trade. Dynamic position capacity scales 0–12 based on VIX and circuit breaker level. 5-level portfolio circuit breaker halts entries at -15% drawdown and liquidates all at -20%. EXIT SYSTEM — 3-LEVEL HYBRID Level 1: Reaasnoing Agent will set the stop loss (-35%/-50%), time stop (7/14 DTE), strategy reversal. Level 2: Exit Guardian agent, activates only at inflection points (PnL ≥ +35% or ≤ -15% or DTE ≤ 21), reducing LLM calls by ~70%. Converts super-runners into dynamic trailing stops persisted to database, targeting +150–300% on winners. Level 3: Deterministic quantitative fallback, zero single point of failure.
4 Sep 2026