Back to digestSYS · OPERATIONAL

/how-it-works

The Daybreak pipeline

A cron job wakes up every morning, pulls a wide net of AI/tech sources, asks Claude to score and structure them under a strict JSON schema, and commits the result back to the repo. The frontend just reads that file. No servers, no database — the pipeline itself is the product.

Days running
0
Items processed
0
Avg / day
0.0
Schema fields
12

Daily run

  1. 01GitHub Actions

    Cron fires every morning at 09:30 ET

    workflow_dispatch + schedule

  2. 02Source pull

    RSS + APIs across ~40 feeds

    dedupe · normalize · truncate

  3. 03Claude

    Structured JSON via tool schema

    temperature 0.2 · schema-locked

  4. 04Commit artifact

    digest.json pushed to main

    signed bot commit · immutable log

  5. 05Frontend

    TanStack Start reads the JSON

    static fetch · no server DB

The whole thing runs in under 90 seconds — most of that is model latency, not IO.

Source of truth: wishmur/tech-digest · digest.json is regenerated in place, so its git history is the archive.