Inbox Triage Pipeline

01 The problem

The inbox ate the first hour of every day. Sorting what mattered from what did not, finding the thread that needed an answer, noticing the deadline buried three messages down. The work happened before the real work even started, and none of it needed a person. Worse, an urgent message could sit unread for hours behind a wall of newsletters and receipts, and I would not know until it was already late.

02 The system

A pipeline that does it overnight. Every evening it pulls the full day of email across every label, drops the obvious noise with plain rules so the expensive step only reads what matters, then hands the rest to a language model that reads each thread and sorts it into four tiers. It writes a ranked briefing and a short action list, then leaves them on a dashboard, in a message, and in a dated archive. By the time I sit down, the sorting is done and the one email that mattered is at the top.

03 How it holds up

It runs on two models, not one. The primary sorts every thread, and a second takes over the moment the first returns anything broken, so the briefing never comes up blank. Three separate layers stop a duplicate run from overwriting a good day with a bad one. It publishes on soft warnings and flags them rather than failing into an empty screen, so I always get a briefing even on an off night. A separate watchdog confirms each night’s run actually landed and raises an alert if it did not, which means the system watches itself instead of waiting for me to notice a gap. 179 tests cover the parts that matter.

04 The result

I got the first hour of every day back. The inbox sorts itself before work starts, so what is left is a short, ranked list of what actually needs a person and a briefing already waiting. I stopped missing the one email that mattered, and the pipeline tells on itself the rare time something slips, so the briefing is always there.