ML Research Engineer · London
I am Albert, a production-minded ML research engineer. My background combines applied modelling with strong MLOps and infrastructure experience, mostly in the life sciences. I am interested in hard technical problems across the wider technology sector.
Selected work
01 / LLM inference research
Poolside Laguna Research Hackathon submission with Cameron Wheeler. Adapted DuoAttention to Laguna's gated attention architecture and used learned head importance for a mixed-precision KV-cache policy, reducing cache size by up to 53%.
02 / Life science tooling
MCP servers that make protein and small-molecule data available to AI agents, with local Ollama-based testing for a practical research workflow.
03 / Transformer fundamentals
A from-scratch Transformer implementation in JAX, Flax and Optax, applied to sentiment analysis to explore attention, training loops and model internals directly.
04 / ML research
An experiment in rotation equivariance: Group Convolution U-Nets evaluated on butterfly segmentation, with tracked model comparisons and a technical write-up.
05 / Model implementation
A from-scratch JAX and Flax implementation of U²-Net for salient-object detection, translating a non-trivial vision architecture into a functional ML stack.
06 / Reinforcement learning
Distributed PPO training for Unity ML-Agents' 3D Ball environment. Uses MPI to run parallel environments and accelerate policy learning.
Technical writing
Longer-form explorations of model behaviour, implementation details and whatever else seems worth understanding properly.
Read the notes →Contact
For ML engineering roles and interesting technical problems:
albertswchung@gmail.com ↗