Samuel Okoye
@samuelAI code review · developer tooling
Getting models to leave useful review comments instead of nitpicking formatting.
on Tutorial: checkpointed agent loops you can safely retry · 1d ago
Same here. Everyone underestimates how much time goes into the boring parts.
on Lessons learned: the five retrieval bugs that reached production · 4d ago
This is the first explanation of backpressure I've seen that isn't a hand-wave.
on Lessons learned running GPU capacity for 14 internal teams · 4d ago
We copied the idempotency-key idea last month and duplicate writes went to zero. Thanks for that.
on Architecture notes: autoscaling GPU inference without melting the bill · 5d ago
Great timing — we're mid-migration.
on Tutorial: hybrid search with pgvector and reciprocal rank fusion · 6d ago
Bookmarking for whenever we finally split this out of the monolith.
on Lessons learned: what 18 months of eval data actually predicts · 11d ago
Careful with that; it bit us when the upstream started returning 200s with empty bodies.
on Tutorial: quantise a model without wrecking its reasoning · 16d ago
Great detail on the failure modes. That section usually doesn't exist. (5)
on Tutorial: quantise a model without wrecking its reasoning · 16d ago
Good question. We never solved it properly either.
on Tutorial: a golden-set eval suite that blocks bad prompt merges · 23d ago
You mentioned 8ms overhead — was that measured client-side or at the edge? (73)
on Lessons learned: 6% of our first LLM outputs were quietly malformed · 25d ago
Thanks for writing down the failure path. Most posts stop at the happy one. (79)
on Lessons learned: 6% of our first LLM outputs were quietly malformed · 25d ago
The 'one shared schema' point is doing a lot of quiet work here. (83)