Resume

AI-native quantamental analytic agent

• Built Financial Deep Research Agent: tool-calling research system integrating Postgres (portfolio), Neo4j (supply chain graph), Qdrant+Voyage finance embeddings (report retrieval), Redis caching, and Exa web intelligence
• Shipped difficulty-tiered evaluation harness to prevent regressions in tool selection, entity grounding, numeric accuracy, and hallucination rate
• Built HedgeLLM: LLM fine-tuned via SFT/RL for delta-neutral hedging; computed Greeks via Black-Scholes across 10M+ daily data points; achieved 85% action accuracy
• Implemented PPO-based RL environment achieving 1.8 Sharpe ratio; backtested across 5 years with <5% max drawdown
• Maintained 95%+ days within ±10 delta of neutral target with multi-component reward function (P&L, delta penalty, transaction costs)

Product development from research-incubated Federated Learning open-source framework to production Generative AI platform.

• Designed 3 product tiers (cloud-hosted, on-premise, hybrid) and GPU pricing for 6 types (A100, H100, T4, L4, RTX 4090, V100); enabled 25+ enterprise customers contributing to 40% of revenue and $500K+ annual compute revenue at 35% gross margin
• Shipped 12+ major features across MLOps, Scheduler, and Compute layers with 8 engineers; reduced model deployment time from 4 hours to 15 minutes through streamlined provisioning workflow
• Led AI Agent platform from PRD to GA in 10 weeks with 4 engineers; launched beta to 150+ users, iterated through 3 versions, achieved 500+ agent deployments in first quarter
• Scaled Model Hub to 30+ open-source models (Llama 2/3, Mistral, Qwen, Falcon, MPT) with <72-hour integration turnaround; served 10K+ daily inference requests at peak

PAI (Platform for AI) powering Qwen model training, deployment, and inference at scale.

• Pioneered the Federated Learning (FL) Privacy module; drove adoption via 10+ PoCs, converting pilots to paid contracts; secured national security compliance approvals, unlocking highly regulated markets
• Led launch of the EAS high-availability module (Blue-Green/Canary deployment, auto-scaling, heterogeneous resource support), achieving 99.9% uptime SLA for MLOps users
• Developed EAS pricing model and billing logic, linking consumption metrics to customer value; improved monetization efficiency and accelerated platform revenue growth

• Engineered CV inference pipeline for ultra-small instrument defect detection (80→25ms latency, 40 FPS on edge); led MLOps deployment and A/B validation ensuring >95% F1 score and reducing model drift by 40%
• Led MLOps deployment and A/B validation of researcher-developed models, ensuring production quality standards (>95% F1 score); implemented a validation framework that reduced post-deployment model drift by 40%

• Focus on Software Engineering, Data Analytics, and Cloud Computing

• Strong foundation in computer science and engineering principles
• Intersection of intelligent system, cryptography, and privacy related topics