Systems engineer · AI infrastructure

Jiahua Wang (Alvin)

I build infrastructure that helps ambitious AI ideas survive contact with reality.

Previously led model evaluation at OpenAI for GPT-4, built training infrastructure for o3, and prototyped agent frameworks. Before that, data platforms at Tesla and backend services at VMware.

01

Selected work

Things I’ve made

022025

chat

A focused Tauri × React desktop client for LLMs: searchable history and secure key storage in a 4MB package.

View project
032025

Guugin Design

A calm, image-led portfolio for an interior design studio, with a responsive editorial layout.

Visit site

02

Experience

Where I’ve worked

2025—Now

Delphina

Early Engineer

Rust feature pipeline. Replaced DuckDB with Rust and built deterministic ref-counting across Ray workers: 2.5–22× faster with 50% less peak memory.

Secure agent runtime. Shipped customer-hosted Kubernetes runtimes using Redis, presigned S3 URLs, and SSE.

2022—2024

OpenAI

Member of Technical Staff

Evals platform. Led the model evaluation system used by 100+ researchers, cutting evaluation cycles by ~50% and providing primary GPT-4 training metrics.

Training & agents. Built o3 experiment infrastructure and prototyped an agent framework used by 200+ employees.

2018—2021

VMware

Software Engineer

Built the Java/Azure backend for VMware’s Azure-native virtual desktop platform.

2017—2018

Tesla

Software Engineer

Developed data pipelines for vehicle, energy, and manufacturing analytics.

03

Toolkit

Tools of the trade

Languages

Python, Rust, TypeScript, Java, Scala, C++, Haskell

Systems & data

Kafka, Redis, PostgreSQL, Ray, ClickHouse

Infrastructure

Docker, Kubernetes, Azure, AWS, GCP, Terraform

Interfaces

React, Tailwind, Tauri