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-
Why I still use Sol on low
Astra surprised me, but GPT 6 Sol on low remains my daily driver. Here's where Fable and Opus 5.5 fit, and why my plan matters.
-
What Does a GPU-Hour Actually Measure?
GPU-hours hide differences in GPU models, whole-device allocations, and MIG slices. Research reports need to say what was actually used and how the total was calculated.
-
Before You Rebuild Your RAG Pipeline, Try an Agent
If RAG is returning incomplete answers from dense technical documentation, try letting an agent search and read the source material directly.
-
Read the incident report before declaring AGI
The OpenAI and Hugging Face incident is a good example of why dramatic claims about AI need to be checked against the full incident report.
-
Your KPIs Aren't Telling You Enough
Uptime percentages and other headline KPIs can hide which capabilities failed, who was affected, and what an organization should do next.