JIAHUA WANG (ALVIN)
Email
GitHub
LinkedIn
Building infrastructure and platforms at the intersection of systems and AI.
Previously led model evaluation at OpenAI for GPT-4 and prototyped agent frameworks. Built data platforms at Tesla and backend services at VMware.
Skills
Experience
Delphina
Early Engineer- Rust × Ray pipeline: Replaced DuckDB with Rust, integrated with higher level coordinator in Ray and implemented deterministic ref-counting to avoid OOMs. 2.5–22× faster, 50% less peak memory.
- Secure agent runtime: Ran in customer data planes via Kubernetes, communicated with web UI via control plane and S3. Used Redis for queuing and notifications, presigned S3 URLs for transmitting agent messages securely, and SSE for streaming.
OpenAI
Member of Technical Staff- Evals platform: Led development of model evaluation system used by 100+ researchers. Cut evaluation cycles by ~50% and provided primary metrics for GPT-4 training.
- Agent framework: Prototyped internal agent system enabling 200+ employees to create experimental agents. Set foundation for public agent products through close collaboration with leadership.
VMware
Software Engineer- Built Java/Azure backend for Hydra, VMware's Azure-native virtual desktop platform.
Tesla
Software Engineer- Developed ETL pipelines (Java, Scala, Hadoop) for vehicle, energy, and manufacturing analytics.
Projects
chat
2025Tauri × React desktop app for interacting with LLMs. Features searchable chat history, smooth scrolling, and secure API key storage in a 4MB package using ~130MB RAM.
View on GitHub →xplog
2025LLM training dashboard rendering 10×10×1M-point charts smoothly on base M3 MacBook. Built with ClickHouse, React, and Python for real-time training metrics visualization.
Guugin Design
2025Interior design firm website with responsive design and minimal aesthetic. Focus on typography and imagery to showcase portfolio work.
Visit site →