Callum Young

Software Engineer at Google, working on agents in Cloud AI.

Founder of olangutan.com.

I come from a background in Computer Science and Applied Statistics. Right now I'm writing my way through a series of deep dives into newer ML topics — trying to build real intuition for the parts of ML I want to work on next. A recurring thread: how agent harnesses orchestrate models for programming, and why some setups produce dramatically better results than others using the same underlying model.

Read the blog →The GOATs →Enter the 3D world →
Callum Young

Learning timeline

Full timeline →

  1. Transformers, end-to-end from scratch

    Re-deriving attention, KV cache, and positional encodings — and what each design choice actually costs.

  2. Harness vs harness: why agent scaffolds differ so much at coding

    Same model, very different results. Comparing planner/executor splits, tool grammars, and memory schemes — and what actually moves the score.

  3. Upcoming

    Speculative decoding and draft models

    How draft-then-verify changes the latency/cost frontier, and where it stops paying off.

  4. Upcoming

    Tool-use traces as a training signal

    What we can and can’t learn from agent trajectories — supervised, preference, and RL angles.