How this works.
Who it's for
One reader: Muaaz Shaikh, a CS student and AI intern who builds agent infrastructure, runs a startup and does security research. The digest picks the handful of stories from the past week that matter to that work and explains each one properly, from an intuition you could tell a friend, up through the architecture, the maths and some code.
The weekly pipeline
- Every Saturday evening a scheduled Claude session starts in the cloud with this repository.
- It reads the reader profile, the source list, the running preference notes and the last few issues, then gathers 25–40 candidate items from the past seven days.
- It scores them against a written rubric (relevance, significance, novelty, depth) and selects about six, with at least one security story and at least one that isn't about AI.
- It reads the primary sources, writes each story in layers (overview first, then the full deep dive), and links every source it used.
- It runs the content checks and a full build. Only a passing build is pushed, and Vercel redeploys the site by about 7 pm IST.
How feedback works
There's no backend. The “Useful / Not useful” buttons on each story and the “Tune next week's issue” chips open a pre-filled issue in the GitHub repository with the feedback label. The next Saturday run reads open feedback issues, folds them into a running preferences file, and closes them with a note.
Accuracy
Every claim should trace to a listed source; quotes are kept short and speculation is marked as such. If something looks wrong, use the feedback button and say so.