---
title: "How it's built"
url: https://namasteji.aiwithayushman.com/build
summary: "The whole system on one page, then eight zooms into it."
type: overview
zoom: 0 (the map)
---

# How it's built — the Build-it track

Namaste Ji is a real, running product: five AI agents research Indian festivals, decide what
to make, generate devotional greeting images, quality-check them, and feed them to an app —
while one human supervises from a dashboard. It runs entirely on Cloudflare. This track walks
through how, top-down: one picture, then eight zooms into it.

## Why it might matter to you

Almost nothing here is specific to greeting cards. The same chassis builds any application
where AI agents produce the product and a human stays in charge: agents that make, a store
they share, a human gate, and a serving path that stays dumb and fast. The last zoom hands
you the chassis as an empty, MIT-licensed repo.

A note on the platform (unsponsored): one Cloudflare account runs all of it, and the
$5/month Workers Paid plan unlocks essentially everything a build like this needs.

## Six words that keep coming up

- **Agent** — a small program with one job: runs on a schedule, thinks with an AI model,
  uses tools, writes results to a database. Not a chatbot.
- **Brief → shot** — a brief is an idea for a batch of content; a shot is one exact picture
  to make, with its words fully written out.
- **The catalog** — the one database of record: rows in D1 (SQLite), images in R2 (object
  storage), similarity in Vectorize (a vector index).
- **Embedding** — numbers describing what a picture looks like, so similarity is computable.
- **The console** — the private dashboard where a human watches runs, approves work, and can
  pause any agent. A read-only copy is public.
- **Slate** — a pre-computed list of ~50 cards waiting in storage for one person's next
  morning.

## Two rules the whole design follows

- **AI decides slowly, code serves fast.** Models think in scheduled background work; the
  moment a person opens the app, everything they touch is deterministic code.
- **A human can stop anything.** Every agent runs inside a gate a person can flip — auto,
  manual, paused — without a deploy.

## The zooms, in order

- zoom 1 — Architecture: https://namasteji.aiwithayushman.com/build/architecture.md — How the system is organized: agents make things, humans oversee, one catalog sits between them.
- zoom 2 — Cloudflare: https://namasteji.aiwithayushman.com/build/cloudflare.md — What it runs on — every service and every connection, drawn from the infra code itself.
- zoom 3 — Agents: https://namasteji.aiwithayushman.com/build/agents.md — What an agent is here, and how one works inside: memory, tools, prompts, an off switch.
- zoom 4 — QA: https://namasteji.aiwithayushman.com/build/qa.md — How quality is controlled — and the failure that made QA a core part of the system.
- zoom 5 — Tooling: https://namasteji.aiwithayushman.com/build/tooling.md — How it is developed and shipped: CI, infrastructure as code, one gateway for every model call.
- zoom 6 — Feed: https://namasteji.aiwithayushman.com/build/feed.md — How each person's feed is put together — and why no AI model sits in that path.
- zoom 7 — Data & analytics: https://namasteji.aiwithayushman.com/build/data.md — What gets measured, how the feed and the content strategy learn from it, and where privacy fits.
- zoom 8 — Starter kit: https://namasteji.aiwithayushman.com/build/starter.md — The same chassis as a public repo — start your own agent-run app from it.

## Navigation

- Next: https://namasteji.aiwithayushman.com/build/architecture.md (zoom 1, Architecture)
- Index of everything machine-readable: https://namasteji.aiwithayushman.com/llms.txt
- If you only read one file: https://namasteji.aiwithayushman.com/AGENTS.md