Jonas Weber
@jonasMCP tooling · DX
Model Context Protocol servers, schemas that models actually follow, and boring installs.
on Tutorial: hybrid search with pgvector and reciprocal rank fusion · 6d ago
We did the boring version of this — a queue and a worker — and it held for two years.
on Tutorial: hybrid search with pgvector and reciprocal rank fusion · 6d ago
Great detail on the failure modes. That section usually doesn't exist.
on Tutorial: hybrid search with pgvector and reciprocal rank fusion · 6d ago
Bookmarked.
on Lessons learned from 11,000 agent runs that went sideways · 8d ago
Honestly the answer for us was fewer features, not better infrastructure.
on Tutorial: build a streaming LLM gateway with per-tenant rate limits · 8d ago
The counter-intuitive part for us was that more retries made availability worse.
on Lessons learned: what 18 months of eval data actually predicts · 10d ago
One gap: nothing here about data residency. That's what killed our multi-provider plan.
on Tutorial: serve an 8B model at 3x throughput with continuous batching · 11d ago
Thanks — that gives me something concrete to try this week.
on Lessons learned: 40k requests a day on a single 24GB card · 13d ago
You mentioned 8ms overhead — was that measured client-side or at the edge?
on Tutorial: quantise a model without wrecking its reasoning · 16d ago
The tuning-knobs table is more useful than most vendor documentation. (20)
on Lessons learned: 6% of our first LLM outputs were quietly malformed · 25d ago
Great detail on the failure modes. That section usually doesn't exist. (88)
on Lessons learned: 6% of our first LLM outputs were quietly malformed · 25d ago
We copied the idempotency-key idea last month and duplicate writes went to zero. Thanks for that. (20)
on Architecture notes: planner–executor–verifier without the framework tax · 27d ago
Curious how you cost this per tenant — token counts only, or wall-clock too? (45)