THE AI RUNTIME/learn

ABOUT

Between the demo and production

The AI Runtime teaches the engineering work between an AI demo and a production system. That gap is not model knowledge. It is what happens when the process dies at step nine, when the publish call times out after it already succeeded, when a reviewer approves three hours later from their phone, and when an auditor asks for the evidence.

The curriculum takes the perspective of forward-deployed engineering: every topic is taught the way an FDE meets it, in order.

  1. Discover Find the real problem behind the customer's ask.
  2. Shape Define scope, constraints and what success measures.
  3. Architect Choose the system shape and the tradeoffs it commits to.
  4. Build Implement with production primitives, not demo shortcuts.
  5. Evaluate Prove improvement and safety with validated evals.
  6. Deploy Ship into the customer's environment and constraints.
  7. Operate Own what happens after failure, at 3am, with evidence.

How this works

  • Free to read and watch. Every course is complete as written and recorded material.
  • Optional open-source implementations. Reference repos exist for people who want their hands on the system; nothing requires cloning them.
  • Optional labs. Failure Labs break the running system on purpose, for those who want to feel it.
  • No account required. Progress, if you want it, lives in your browser and nowhere else.
  • Not a paid LMS. No paywall, no seat licenses, no completion certificates.
  • Curated external resources. When another source teaches a prerequisite better than we could, we link it and say why, official documentation first.

Who is behind this

Written by Kranthi Manchikanti, a solutions architect working on agentic systems in regulated and healthcare settings, and the person behind The AI Runtime: a publication, a Boston meetup, the FDE Lab, and the FDE Talks podcast.

Everything here is public and reproducible on purpose: deterministic fixtures, pinned environments, open datasets and evaluators. Claims about reliability are only worth something if a stranger can re-run them.

Contact

For production reviews, custom eval systems, failure post-mortems, or private cohorts, reach out. Quality over noise is the filter.