Tyler Collins
Research computing consultant, Canada
I help researchers turn computational problems into reliable software and workflows: HPC systems, Python tooling, EEG/MEG pipelines, getting more out of the hardware you already have, and helping scientific teams use AI coding agents well.
Recent writing
All writing-
Send your agent a screenshot
Agents only know the dashboards they were trained on. When menus keep moving or a UI looks wrong, a screenshot gives the agent what's actually on your screen.
-
Get a second opinion from another model
Models are tuned to write code differently. If you have the tokens, having one review another's work is basic hygiene, and personas take the idea further.
-
Clean-Room Rewrites for Researchers
Tools like REA let a coding agent take closed software apart. Escaping research lock-in is one of the best uses for them, as long as you stay in your lane.
-
Writing an AGENTS.md for a national cluster
What I learned putting an AGENTS.md into the Alliance software stack: skip what frontier models already know and spend the lines on site policy and when to stop and ask.
-
Your LLM Loop Is a Batch Job
800,000 small requests in a day slows a shared LLM service down for everyone. If your workload looks like that, a batch job on a GPU node will finish sooner.