Hello,

I'm Chang Min, a builder and engineer at Meta interested in ML infrastructure and distributed systems.

Most recently at Meta, where I took database platform SLA availability to 99.99% for 125+ enterprise customers. Currently contributing to SGLang and Kubernetes on LLM inference performance.

Chang Min holding a coffee in front of Kopiku, an Indonesian café in San Francisco.
About

Curious about systems that scale.

I like the messy middle of software: the place where a request leaves one machine, gets queued, balanced, replicated, and still has to come back fast. That's pulled me toward databases, backend infra, and lately the emerging problem space of LLM inference at scale.

I'm also drawn to the founder side of building: the why behind what gets built. I've shipped end-to-end at an early-stage startup (Hangry), competed in the YC × Google DeepMind hackathon, and run my own products. Early-stage technical tradeoffs (what to build, what to skip, what to bet on) are the kind of questions I like chewing on.

I'm also a passionate researcher. I have done CNN research with Professor Thiago Serra and I am currently exploring ideas around reinforcement learning and continual learning with Professor Brian King.

Outside of work, I work on open source (SGLang, Kubernetes SIG Inference-Perf, and an AI Gateway / envoy project), read papers on inference serving, and occasionally ship a side project end-to-end.

Education

Education.

May 2026

Bucknell University

B.S. in Computer Science and Engineering · Lewisburg, PA

  • Relevant coursework: AI with Neural Nets, Operating Systems, Computer Systems, Computer Networks & Security, Algorithms Design & Analysis, Software Engineering & Design, Data Structures & Algorithms, Linear Algebra, Discrete Structures, Entrepreneurial Finance.
Work

Experience.

May – Aug 2025

Meta

Software Engineering Intern, Production Engineering · Fremont, CA

  • Took platform SLA availability to 99.99% across MySQL, MSSQL, and MongoDB for 125+ enterprise customers via an automated alerting system built in Python and Chef.
  • Cut SEV recovery time by 67% (15→5 min) with custom observability dashboards for the distributed database fleet.
  • Added 400+ new database metrics and logs, extending the collection framework and reducing operational overhead across multi-datacenter deployments.
  • Automated mass failover validation for 500+ HA databases via a new Thrift API and CLI subcommand surfacing real-time cluster status.
May – Aug 2024

Hangry Indonesia

Backend Engineer Intern · Remote / Jakarta

  • Optimized the notification microservice by 97% (30ms→1ms) by rewriting it in Go with concurrent, async workflows, serving 20,000+ DAU.
  • Reduced cloud infrastructure cost by 70% by replacing RabbitMQ with a Redis Streams queue, connecting PostgreSQL directly on AWS, and tuning queries.
  • Shrank finance audit time by 75% (4 days→1 day) with a containerized supply-order verification service using GPT-4o image processing over REST.
Open Source & Projects

Things I've been building.

Feb 2026 – Present

SGLang

Open Source Contributor · sgl-project/sglang →

  • Raised LLM inference throughput by 20% on Apple Metal devices by enabling CPU–GPU overlap scheduling via asynchronous evaluation of MLX arrays.
Oct 2025 – Present

Kubernetes SIG Inference-Perf

Kubernetes Member / Open Source Contributor · kubernetes-sigs/inference-perf →

  • Extended the benchmarking framework with MultiLoRA support, enabling simultaneous evaluation of 300+ adapters on OpenAI-compatible servers (vLLM); surfaced 10% latency variance across configurations.
  • Architected a high-concurrency load generator to simulate 100+ concurrent clients for LLM benchmarks in Kubernetes.
Jan – May 2025

Chiron

Personal Project · Chiron-AI →

  • Shipped a production RAG pipeline on FastAPI with semantic embeddings over 7+ file formats (PDF, PPT, DOCX) via the docling parser and Qdrant for retrieval.
Also on GitHub

More

envoy AI Gateway · Lantern (neural net training engine) · Car Mechanic AI Assistant (YC + DeepMind hackathon) · PR Review AI Agent · TCP server in C · containers in Go · Google Summer of Code (MIT App Inventor / Google Blockly) · GitHub →

Stack

Tools I reach for.

Languages

Python, Go, Rust, C, C++, TypeScript, Java, SQL

Distributed Systems

Kubernetes, Docker, gRPC, Apache Thrift, Envoy, consensus (Raft), high availability

ML Infrastructure

SGLang, vLLM, llm-d, AI Gateway, Kubernetes inference performance, inference serving

AI/ML

PyTorch, Apple MLX, Weights and Biases, Torchmetrics, Reinforcement Learning

Backend & Data

FastAPI, Node.js, REST, GraphQL, MySQL, PostgreSQL, MongoDB, Redis, Qdrant, message queues, microservices

Devops & Dev Tools

CI/CD, Infrastructure-as-Code (Chef), AWS, UNIX (Linux, MacOS), GitHub, Prometheus, Claude Code

Contact

Say hi.

Always happy to talk about ML infrastructure, distributed systems, or a project you're building.

/in/chang-min-bark LinkedIn @changminbark GitHub @changmin_bark X