Skip to content
Mohamed Khelifa Zaghdidi I lead engineering teams, I run a model, I publish the measurements.

Local inference, measurement, decision

I lead software engineering teams in health insurance. Alongside that, I run an inference system I built and instrumented myself, and I publish what it teaches me — measurements and gaps alike.

The register, in brief

everything measured is written by hand. so is everything missing.

Cache reuse, exact repetition99.9 %
Cache reuse, distinct registers0.06 %
Duration / candidate count correlationr = 0.999
Memory bandwidth, measured ceiling41 GB/s
Share of documents with a reconstituted text layer, ground-truth corpusnot measured
Actual unit costnot measured

readings taken 12 August to 2 September 2026 — the full register

Recent notes

  • Data — September 10 The failure that raises no error

    A scanned document produces text, so a pipeline that expects text works. It also works when it has read only part of the document, and it says nothing.

  • Self-hosted inference — September 8 It is almost never the model

    One hundred and nine missed statements, five mechanisms, and the model is behind none of them. The vocabulary gap everyone assumes is dominant turns out to be the smallest category.

  • Managing agents — September 4 An agent never asks what I mean

    I lead engineers by day and an organization of agents by night. The difference is not speed. It is that the cost of framing no longer amortizes.

All notes — RSS feed