Mufeez Amjad

I'm an engineer based in NYC 🗽, working at Linear, where I predominantly work on infrastructure/systems problems.

On nights & weekends, I'm currently exploring reinforcement learning science & misalignment in multi-agent settings.

Previously, I've worked at:

Meta on Buck2 (Rust)
Meta's incremental build system
, improving build observability across the majority of Meta's CI/CD
Cockroach Labs on Pebble (Go)
RocksDB inspired KV store, powering CockroachDB
, an LSM-tree storage engine handling petabytes of data — writeups: [1] [2]
Splunk on the Release Train (Go)
Programmatically coordinating releases across Splunk's production fleet
team, automating rollouts across a Kubernetes fleet of 10,000s of nodes

I'm a UWaterloo Software Engineering

graduate, and I'm interested in: high-performance computing, distributed systems, and ML systems.

Feel free to reach out on Twitter or over email – I'm always open to chat or grab a coffee!

Training an LLM to use a Debugger

project

Making Qwen3-Coder-30B-A3B better at debugging with SFT and RL

Manage multiple AI agents like Claude Code and Codex

Taking an NVFP4 Grouped GEMM kernel from 238 µs to 23.8 µs with TMA, warp specialization, and cluster multicast.