← compare
Petros Stergioulas

athens · 2026·04 · v05

// entry

26·04·24 · 14:02 EEST

  • file hello.md
  • focus platform · evals
  • status open to talk

I keep a quiet journal of the systems I work on.

I'm Petros — a software engineer who writes things down. This site is the indexed version of a small private notebook: notes that survived, projects that taught me something, and decisions I'd still defend. Nothing here is performance. It's just the readable parts of how I work.

If you're looking for the resume, it's still on LinkedIn. If you're looking for how I think, the index is below.

// section

01 · now

A short list. Updated when something actually changes.

Bearings, end of April.

  • focus

    Reducing alert noise on a platform team without losing the signal.

  • reading

    Naur's Programming as Theory Building — for the third time, slower.

  • writing

    A small CLI to keep team decisions linkable across years.

  • listening

    Mostly Greek piano music. It pairs well with diff-heavy afternoons.

// section

02 · notes

Dated, plain, mostly short. No teaser cards.

Recent notes.

  1. 2026·04·24

    On second-best decisions

    A note on why the decision a team can keep agreeing with usually beats the optimal one. With three small examples, none of them flattering.

  2. 2026·04·17

    Three signals are usually enough

    Saturation, latency, and error budget — done well — beat a dashboard wall. What we removed and what survived.

  3. 2026·04·09

    A week on the wrong abstraction

    Spent a week with a generic adapter that became a junk drawer. What I should have asked first.

  4. 2026·02·11

    Why I keep redrawing the same diagram

    A short essay on how the same loop keeps reappearing across projects, and why I've stopped trying to break out of it.

archive · 2024–2026 →

// section

03 · projects

A few records of work, with the trade-offs left in.

Selected projects.

  1. 2024 — present · platform

    Quiet observability

    Replaced a noisy dashboard culture with three durable signals and a runbook. Pages dropped 64%. The hard part wasn't the metrics — it was writing the runbook in a way the on-call could read at 3 a.m.

    trade-off lost some operational variety, gained a team that sleeps.

  2. 2023 · api

    Boring versioning, on purpose

    A small versioning policy nobody complained about. The clearest writing was the most contested PR; the implementation was a weekend.

    trade-off slower change, fewer surprises for downstream teams.

  3. 2022 · internal

    A tiny eval harness

    Made an opinionated team agree on what "better" meant for two months. Then we forgot, and that turned out to be fine.

    trade-off less rigor, more shipping. We picked shipping, and re-introduced rigor selectively.

// section

04 · records

A small log, the way it's actually kept on disk.

The journal, viewed plainly.

Every entry on the site is a flat Markdown file in a private notebook. Here's what the latest log looks like — date, slug, status, and the small shape of the change. 

$ git log --oneline -n 87d3a1c notes/2026-04-24-second-best-decisions.md +helda01fb2 notes/2026-04-17-three-signals.md +heldce442d notes/2026-04-09-wrong-abstraction.md +draft → +held22ff9e decisions/dec-011-resume-tools.md +heldbd0017 projects/quiet-observability.md +update5e1c83 notes/2026-02-11-redrawing-diagram.md +held8aaf04 decisions/dec-009-thirty-min-reviews.md +retired12e6b1 notes/2025-12-20-runbook-nobody-reads.md +held

The site is just an index over this folder. If a note isn't here, it isn't yet.

// section

05 · contact

Plain channels. I read all of them.

Open ways to reach me.