Hi!
š Iām Karanjot Vendal, an ML Engineer based in London.
Iām currently exploring the space between research and engineering for large language models ā understanding transformers from the math to the metal, and building things that run fast.
This site is a public record of that work:
- Weekly: a running log of what I study, build, and get wrong each week.
- Posts: deeper writeups on specific concepts or projects.
- Projects: code, experiments, and reproductions.
Previously, I worked on sports simulation and real-time ML at Hawk-Eye Innovations, and I hold an MSc in AI from Queen Mary University of London.
Weekly
29 Jun 2026
June 29 ā July 5, 2026
This week #
- Revived this site as a public record of a 6-month research engineering trajectory.
- Settled on a structure: transformers + LLM inference + efficient kernels as the core trunk; agents, SOTA papers, and TypeScript literacy as satellites.
- Blocked study time on the calendar for the next 8 weeks.
- Started a living note to track direction: Research Engineer Trajectory.
Up next #
- Read the Word2Vec paper and understand CBOW / skip-gram / negative sampling.
- Implement Word2Vec from scratch in NumPy or raw PyTorch.
- Explore Tau as a Python agent harness for MVF work.
Open questions #
- What part of the transformer/inference stack lights me up most as I go deeper?
- How much focused work can I realistically get from the 3ā7pm weekday blocks?