Atlas

Live operations dashboard — from spreadsheets and phone calls to one shared picture the team used every day.

Problem

A regional operator ran a growing network on spreadsheets, phone calls, and tribal knowledge. Decisions rode on data that was hours old; nobody could see the whole picture at once.

Role

Lead engineer — event pipeline, state model, and operator-facing dashboard. Worked closely with people who lived in the old process.

Approach

Started with the single worst blind spot (live status), then layered capability only after the previous one was trusted in daily use. No big-bang cutover.

Architecture

Kafka ingestion → Postgres materialised state → React + map UI. Named everything in the operator’s vocabulary, not the engineer’s.

Outcome

Ran in parallel with the old process until the team trusted it enough to let the spreadsheets go. Cutover was deliberately boring.

Lessons

The hard part wasn’t the software — it was earning the right to be the source of truth.