
Small Time Devs builds Aramid: a quantitative trading-intelligence platform where deep-learning forecasting, a vision-language trading copilot, and autonomous strategy research work as one system — GPU-accelerated end to end.
Aramid isn't a collection of indicators — it's an integrated intelligence stack where every layer, from raw market microstructure to the voice in the trader's ear, is a model we built, trained, and validated.
GRU sequence models forecast intraday range, volatility, and tail risk — outperforming the HAR-RV econometric benchmark by 15–26% out-of-sample. Trained and served on our own GPU fleet.
Aramid Chart Coach reads a live trading chart through a locally-hosted vision-language model, fuses it with authoritative market data, and coaches the trader in real time — by voice. Applied VLM, in production.
A locally-hosted large language model turns dense market structure — volatility regimes, options positioning, liquidity — into institutional-grade plain-English interpretation, live on every surface.
Meta-labeling models score every trade candidate's probability of success and size it accordingly — a validated risk-adjusted overlay that concentrates capital into the highest-quality opportunities.
A 24/7 research engine searches a vast strategy space with genetic algorithms, then subjects every candidate to a rigorous statistical gauntlet. Machines propose; statistics dispose.
Nothing ships on a promising backtest. Every strategy must survive realizable-fill modeling, multiple null-hypothesis tests, out-of-sample confirmation, and adversarial review — then prove itself forward.
The platform ships as a professional trading terminal — live market structure, model forecasts, AI narrative, and the vision-language Chart Coach, delivered to serious futures traders as a subscription product.
Real-time options-positioning structure, deep-learning range forecasts, AI-generated market narrative, and an on-chart copilot — the full stack, live on web and in the Quantower professional platform.
aramidtrading.com →Markets punish wishful thinking. Every strategy candidate — human-designed or machine-discovered — must survive realizable-fill simulation, multiple independent null-hypothesis tests, out-of-sample confirmation, and adversarial review before it earns forward capital. We publish nothing to our own trading desk that hasn't beaten its own null. The result is a small book of validated strategies — and a large, honest graveyard.
We're a product-focused engineering company applying modern machine intelligence — deep learning, vision-language models, large language models, and evolutionary search — to one of the hardest real-time problems there is: understanding and trading live markets.
Small is the point. A tight team, our own GPU metal, our own models, and a discipline of proving everything — shipping an intelligence platform that institutional desks would recognize, at a scale where every component is understood end to end.