Independent Algorithmic Trader · Real-Time Financial-AI Systems
I trade for a living and build the systems I trade with — real-time 0DTE SPX options-microstructure infrastructure, and now a self-hosted AI layer that learns my methodology and reasons over live market structure.
What I do
Defined-risk options strategies on SPX and index products, driven by dealer positioning and regime. My trading systems carry strong Calmar and Sortino ratios across forward testing and live management — trading is my primary livelihood.
Tick-level ingestion and per-minute aggregation of gamma / delta / vega exposure, charm, skew, max-pain and net-premium drift across the full 0DTE SPX chain, VIX and 250+ concurrent instruments.
Local LLM agents that learn the way I trade — from my notes, daily write-ups and past sessions — through a pgvector memory layer, pre-warmed with dense structural summaries instead of raw ticks.
A strategy registry with walk-forward validation and prop-firm drawdown/survival modeling, with reproducible result caching for honest, comparable evaluation.
Where I'm headed
Today I rotate one-shot calls across commercial APIs — Claude (the most capable), plus Gemini, Mistral and others — feeding each a tightly compressed market summary, because tick-level data won't fit and the plan caps run out mid-week. Moving to a self-hosted model removes the cap and lets me benchmark what actually matters for trading:
Time-to-first-token and generation latency low enough to fit inside scalping execution windows.
Sustained, minute-by-minute context injection through the session without thermal or VRAM bottlenecks.
Rapidly experimenting to learn which real-time signals carry edge — and which are noise — across high volumes of structured market data.
Decision speed fast enough for the model to react inside scalping windows — quicker than manual execution of the same methodology.
The longer-term goal is a refined local model with durable long-term memory — warmed from summary profiles and continuously taught my methodology — running at high throughput on dedicated hardware.
Selected work
A real-time SPX options analytics and automated-reporting platform: per-minute dealer-exposure snapshots, filtered net-premium drift, regime detection, and gated end-of-day / pre-open / intraday report generation — surfaced through a React charting portal, with verification gates on every published figure.
Multi-agent orchestration over a knowledge base of thousands of embedded research chunks with structured tool use. It learns from my daily write-ups and past sessions via pgvector — and is being moved off API caps onto self-hosted hardware at near-zero marginal cost per token.
Backtesting, a strategy registry and signal research (dealer-positioning, regime and ML families) with prop-firm survival simulation and reproducible, cached result sets for comparable evaluation.
Daily intelligence
Every session my stack generates an end-of-day market report — dealer-positioning, regime, flow and the levels that matter — fronted by $RAVOLM, a regime-aware animated volume visualisation. A few recent ones:
2026-06-04 · Thursday
Opened into the 7,500 fears and printed the low minutes after the bell, then ran one-way higher as dealer positioning flipped bullish — tagging the 7,600 call wall before fading into the close.
2026-06-03 · Wednesday
A one-way price-down session that closed under the gamma flip and pressed after-hours toward a descending max-pain target — with the bullish closing drift framed as a session footprint, not a forecast.
2026-06-02 · Tuesday
Strongly one-way put-selling drift under a positive-gamma cap, with a dealer-long delta shelf supporting price and a descending max-pain level as the lone counterweight.
$RAVOLM (Regime-Aware Animated VOLM) — research/visualisation, not financial advice.
Curriculum vitae
Former CISO and CTO with 20+ years in regulated financial services and digital security — now an independent algorithmic trader and full-stack systems developer. I design, build and operate the systems I trade with: real-time 0DTE SPX options analytics, dealer-exposure and regime engines, and a multi-agent AI / reporting layer with a pgvector memory. My trading systems carry strong Calmar and Sortino ratios across forward testing and live management, and I'm now building toward self-hosted LLM inference to apply local models to my methodology at high throughput.