DesignDevelopDeploy

Jan 2025 - Present

• •Saved $10K+ annually by building a cross-platform build pipeline for the OEMS C++ trading engine using Conan + OSXCross, eliminating developer VMs and enforcing deterministic builds across architectures.
• •Deployed and served Qwen MoE 27B on on-prem multi-GPU infrastructure using vLLM, enabling low-latency inference workflows for internal developers and traders.
• •Built a production-grade GenAI trading copilot enabling natural-language workflows across 100+ trading APIs using RAG pipelines, semantic reranking, and human-in-the-loop systems.
• •Built post-trade analytics infrastructure with real-time ingestion and reconciliation pipelines enabling slippage analysis, execution validation, and mismatch detection across exchanges.
• •Enabled multi-exchange trading support by integrating margin and leverage APIs while handling exchange-specific edge cases and execution nuances.
• •Engineered high-frequency systems handling 300+ websocket messages/min using efficient rendering and bounded queues to eliminate lag and memory issues.
• •Reduced symbol registry latency by 60% (5s → 2s) for 3,000+ assets/exchange using Redis rolling caches and pagination for latency-sensitive systems.