Case Study
Full-Stack Low-Latency Market-Making System
Architected and deployed a proprietary HFT market-making system from scratch, operating live on Binance and OKX with a ~12 ms hot-path execution engine and cross-venue options hedging on Deribit.
Challenge
Polykern needed a complete proprietary algorithmic trading stack built from the ground up — spanning market data ingestion, strategy research, execution, and risk management. The system had to operate continuously on live capital across BTCUSDT and ETHUSDT with strict uptime and latency requirements, while maintaining inventory neutrality across volatile market conditions.
Solution
Helios architected the full stack: a fault-tolerant data ingestion pipeline at 10–100 ms latency across 8+ exchanges, a ~12 ms hot-path trading engine with performance-critical components in C++ and Python, and an inventory-driven market-making algorithm with real-time quote skewing and exposure control. Cross-venue hedging was implemented via Deribit options to manage directional and volatility exposure. Partitioned PostgreSQL with automated S3 archival handled market data at scale, and distributed PySpark pipelines supported model retraining on terabyte-scale datasets. FIX protocol connectivity and colocation work was handled directly with exchange teams.
Technical Responsibilities
- Architected and developed a full-stack low-latency market-making system from scratch, covering data ingestion, research, and execution layers
- Designed a high-frequency market-making algorithm with ultra-low latency execution and predictive modelling for optimal queue position and execution priority in order books
- Built strategy logic with inventory risk management — dynamically adjusting quotes and exposure based on real-time market conditions
- Deployed and managed trading strategies on BTCUSDT and ETHUSDT across Binance and OKX with cross-venue execution optimisation
- Implemented cross-venue hedging strategies using Deribit options to manage directional and volatility exposure
- Built fault-tolerant high-frequency data ingestion pipelines (10–100 ms) across 8+ exchanges — production incidents as infrequent as ~90 days
- Designed partitioned PostgreSQL storage with automated archival to S3 for robust, high-performance querying of large-scale market data
- Developed a low-latency trading engine (~12 ms hot path) with performance-critical components in C++ and Python for deterministic execution
- Engineered systems supporting continuous strategy execution with high uptime guarantees under volatile market conditions
- Built distributed model training pipelines using PySpark on terabyte-scale datasets
- Designed and implemented Smart Order Routing to minimise slippage and market impact
- Integrated FIX protocol connectivity and collaborated directly with exchange teams for production-grade deployment and colocation setup
- Led latency optimisation initiatives including infrastructure placement, network tuning, and exchange-level benchmarking
Outcomes
- Live HFT system operating on real capital across Binance and OKX
- ~12 ms hot-path execution engine with production incidents as infrequent as ~90 days
- Cross-venue options hedging on Deribit controlling directional and volatility exposure
- Terabyte-scale data infrastructure supporting continuous model retraining and research
- FIX-connected, colocation-ready production infrastructure