Looking for a job

Hao Yang

@haoyang1

Message

I build production AI inference, MLOps, and quantitative trading systems.

United States

Message

What I'm looking for

I'm looking to build production AI, MLOps, or quantitative systems where I can own work from data and infrastructure through reliable, measurable deployment.

I deployed five Qwen3-family LLM inference services at TSMC on Kubernetes, using NVIDIA Run:ai, NVIDIA Dynamo, and vLLM to expose OpenAI-compatible APIs for validation and benchmarking.

I configured disaggregated prefill/decode serving and benchmarked inference topologies under 17.6K-token, 32-concurrency workloads. KV-aware routing improved throughput by 46.4% and reduced average time to first token by 49.2%.

At Medina Partners, I brought research trading algorithms into production on AWS, from market-data ingestion through live order execution. I also built a three-tier data warehouse, automated daily ETL for roughly 2,000 securities, and delivered a React and Django portfolio dashboard with live strategy P&L visibility.

In my projects, I shipped an MLOps music recommendation system with 0.80 production AUC and 13ms p95 serving latency, and built an adaptive INT8 KV cache that reduced Mistral-7B memory use by 58.5%.

Experience

Work history, roles, and key accomplishments

AI Service Engineer Intern

Taiwan Semiconductor Manufacturing Company

Jul 2026 - Aug 2026 (1 month)

Deployed five Qwen3-family PoC LLM inference services on Kubernetes using NVIDIA Run:ai, NVIDIA Dynamo, and vLLM, exposing OpenAI-compatible API endpoints for internal validation and benchmarking. Configured and deployed disaggregated prefill/decode serving on Dynamo, tuning per-role GPU allocation and replica counts, and configuring TCP-based KV-cache transfer between stages.

Kubernetes Run:Ai NVIDIA Dynamo VLLM Qwen 3 LLM Inference Disaggregated Serving KV Caching AIPerf Benchmarking

Quantitative Developer

Medina Partners

Mar 2024 - May 2025 (1 year 2 months)

Developed and deployed trading systems on AWS, engineering the research team's trading algorithms into production, from data ingestion to live order execution. Designed a three-tier data warehouse with an automated daily ETL pipeline ingesting market data across ~2,000 listed securities, and built a full-stack internal portfolio dashboard with React and Django.