Home/ Curriculum/ Day 9
DAY 09 of 12

AI plus ops

Make the site think — and make sure you find out when it's down

You can do… by end of day
  • Run an AI as another HTTP service (llama-server) or call a hosted one (OpenRouter).
  • Structure a prompt to return clean JSON your app can parse.
  • Set up uptime monitoring and basic alerting (Uptime Kuma / Prometheus / Grafana).
  • Parse and act on logs; stand up a private encrypted network (WireGuard/ZeroTier).
Deliverable

A site with a working AI feature + an uptime monitor that alerts you when it's down.

Big-picture hook

It can think, and it tells you when it's sick. One rapid-fire day of the next rungs, then we scope your capstone.

Concepts what you'll understand
  • AI as a service: a model is just an HTTP endpoint — send a prompt, get text/JSON back. Run it yourself or rent it.
  • Prompting for JSON: ask for a strict shape, parse it defensively, and handle the model being wrong.
  • Observability: metrics, logs, uptime; a health endpoint; a simple alert that pages you when it's down.
  • Private networks: WireGuard / ZeroTier — an encrypted tunnel between you and your servers, for ops access.
The lab hands-on, ends with something live
  1. First AI call: send a prompt to a local or hosted model and print the reply.
  2. JSON prompt: get the model to return a strict JSON object and parse it in Python.
  3. Health + logs: add a /health endpoint, wire an uptime check, and parse your app's logs for errors.
  4. Private net: configure a WireGuard peer so your laptop reaches the box over an encrypted tunnel.
Want this in another format? All formats