A live derivatives trading platform for Deribit — designed, built, and operated in production. Every component of our client systems is drawn from this foundation.
This platform manages the full lifecycle of options trading on Deribit — from live market data ingestion and real-time volatility surface construction, through risk management and order execution, to portfolio monitoring and decision support. A self-supervised ML pipeline runs in parallel, generating directional signals on the ETH perpetual from hourly market data. It runs continuously in production and has been refined through live trading.
The architecture is deliberately modular: each component has a single responsibility, communicates through well-defined interfaces, and can be replaced or extended independently. This is our core design philosophy.
Maintains a persistent WebSocket connection to Deribit. Normalises the full instrument universe — options, futures, perpetuals — and publishes order book snapshots, trades, and user events onto the Redis bus. Handles reconnection with exponential backoff.
Calibrates SVI (Stochastic Volatility Inspired) volatility surfaces every 5 seconds across all active expirations. Produces parametrised skew metrics — ATM implied volatility, risk reversal, butterfly spread — and publishes them for downstream consumption.
Computes live portfolio Greeks — delta, gamma, vega, theta — adjusted for volatility skew. Tracks net exposure across all positions, enforces risk limits, and exposes a REST pricing API. Integrates quanto adjustment for USD P&L on ETH-denominated options.
Manages order lifecycle end-to-end. Supports automated delta hedging via perpetual rebalancing, implied volatility-targeted limit orders, microstructure-driven strategies using order book imbalance signals, and manual order submission via the operator interface.
A real-time web dashboard streaming live position data, Greeks, vol surface parameters, and order book state via WebSocket. Provides a single-screen view of system health and market exposure for the operations team.
A PydanticAI-powered Telegram agent with access to the full system via an MCP server (QuestDB, Redis, order submission). Runs four concurrent scheduled loops: hourly volatility reports, economic calendar digest, live signal forwarding from strategyx, and hourly market data ingestion into QuestDB.
Hosts a self-supervised ML models. Serves inference via REST at runtime. Trained offline using walk-forward cross-validation with Sharpe loss.
Evaluates ML strategy at regular time intervals. On each new bar constructs the feature window, calls calibratorx for inference, and — when signal strength exceeds threshold — publishes a trade signal (direction, entry, take-profit, stop-loss) to the Redis signal stream.
Underpinning the system is a proprietary Rust quantitative library developed in-house — the same library deployed across all client engagements.
Black-Scholes-Merton (European, American, barrier, digital, spread), Heston stochastic volatility, Rough Heston with Volterra kernels.
Raw SVI parametrisation with vega-weighted grid-search calibration. Sub-second fit across 100+ strikes per expiry. Full skew metrics output.
Delta, gamma, vega, theta, vanna, volga — skew-adjusted and quanto-corrected for multi-currency derivatives books.
Particle swarm optimisation, simulated annealing, genetic algorithm, and Armijo-backtracked projected gradient for multi-bet Kelly criterion.
Order book imbalance via EWMA smoothing, cumulative delta detection, tick and volume imbalance indicators for short-horizon signal generation.
JEPA masked-patch transformer encoder for latent market representation learning. PositionMLP trained on embeddings with Sharpe loss. Walk-forward validation with ATR-based take-profit and stop-loss brackets.
Marimo notebooks for signal research connected to live QuestDB data: GARCH volatility forecasting, HAR realised vol modelling, SSVI surface fitting, and fractional differencing for stationarity-preserving feature engineering.
Live screenshots from the production system.
Risk Monitoring Dashboard
Real-time Skew adjusted Greeks
Volatility Surface
Real-time SVI calibration per expiry
Market Positioning
Open-interest option tracking
LLM Powered System Chatbot
Market/position monitoring or order execution in natural language
Every component of this system is available as a foundation for a bespoke client engagement. We adapt and extend it to fit your market, your workflow, and your regulatory environment.
Discuss Your Requirements