RecipientOFQuotes

The heart of our trading platform. Real-time quote collection, technical analysis, RL inference, and order execution.

RecipientOFQuotes
Collects real-time market quotes via WebSocket from Binance, Bybit, and StonFi. Calculates technical indicators (RSI, MACD, Bollinger Bands, ATR) in zero-copy pipelines. Runs RL models to generate forecasts and executes trades using Kelly Criterion position sizing. All without human intervention.
Go 1.23+ WebSocket Kelly Criterion RL Inference Real-time

RecipientOFQuotes-Worker

Off-chain optimization and analytics. Genetic algorithms, strategy A/B testing, and model training.

Worker
Runs genetic algorithm optimization to continuously tune strategy parameters. A/B testing of competing strategies in production shadow mode. Meta Learning and Transfer Learning cycles to adapt models across market regimes. All analytics stored in ClickHouse for historical analysis.
Genetic Algorithm A/B Testing Meta Learning ClickHouse

Gorgonia Service

Neural network price prediction. Go-native deep learning without Python bridges.

Gorgonia Neural Service
Neural network price prediction service built on the Gorgonia library. Model versioning with ClickHouse-backed storage. REST and WebSocket APIs for inference. GPU-aware memory management for batch prediction. Models are versioned, monitored, and automatically rolled back on degradation.
Gorgonia Deep Learning REST API WebSocket GPU

StonFi Emulator

Safe strategy validation. Full TON blockchain simulation without mainnet risk.

StonFi Emulator
A full-featured, mathematically accurate emulator of the StonFi DEX on the TON blockchain. Used by agents and researchers for safe strategy testing before production deployment. Simulates order book dynamics, slippage, fee structures, and network latency. Every strategy passes through the emulator before touching real capital.
TON StonFi DEX Simulation Strategy Validation

model-ML Library

The internal Go library that powers every ML component. Shared code, not duplicated effort.

LF

Lock-Free Queues

SPSC and MPMC queue implementations with zero-allocation hot paths. Backbone of all inter-goroutine communication.

CA

Adaptive Caching

LRU, LFU, and ARC with lock-free reads. Self-tuning based on access patterns. Critical for market data replay.

DQ

Q-Learning

Deep Q-Networks with experience replay, target networks, and double Q-learning. GPU-aware memory management.