Production AI Engineer

What you'll build

• 01A retrieval system with measurable groundedness and a CI eval harness that fails the build on regression.
• 02A tool-using agent loop with budget limits, structured logging, and a runbook a teammate can follow at 2am.
• 03A model-serving deployment behind a real load balancer, with rollout, rollback, and cost telemetry.
• 04A capstone repository — code, evals, runbook, postmortem — that the public ledger links to from your credential.

Cohort schedule

Cohort opens

Wednesday, 3 June 2026

Enrolment closes

Wednesday, 20 May 2026

Capstone due

2026-09-09 (week 14)

Cadence

Two 90-min live sessions per week + one office hour

Timezones

Cohort runs in two timezone bands: Americas/EU and EU/APAC

1. Week 01

   Eval-first development

   Artifact Golden-set v1 + a failing CI test that passes locally

2. Week 02

   Retrieval that survives review

   Artifact Indexed corpus + groundedness scorer wired to CI

3. Week 03

   Tool-use loops without runaway cost

   Artifact Agent loop with budget limit, audit log, replay harness

4. Week 04

   Structured logging and audit trails

   Artifact OpenTelemetry trace export from a live request path

5. Week 05

   Mid-cohort calibration review

   Artifact 1:1 mentor review of evals + decision log signed off

6. Week 06

   Model serving and rollout strategy

   Artifact Canary rollout plan + automated rollback runbook

7. Week 07

   Cost, latency, and capacity planning

   Artifact Per-request cost dashboard + SLO doc

8. Week 08

   Safety, refusal policy, abuse handling

   Artifact Refusal taxonomy + red-team report

9. Week 09

   Postmortem culture and on-call

   Artifact Written postmortem on a synthetic incident

10. Week 10

    Capstone scoping and acceptance criteria

    Artifact Capstone spec accepted by mentor and a peer reviewer

11. Week 11

    Capstone build · part 1

    Artifact Working v1 with eval harness on main branch

12. Week 12

    Capstone build · part 2

    Artifact Deployed v1 with monitoring + a published runbook

13. Week 13

    External review

    Artifact Two external engineers review the repo against rubric

14. Week 14

    Capstone defence and credential issuance

    Artifact Signed W3C VC + indexed entry in the public ledger

• Anika Rao

  Staff AI Engineer · ex-Stripe Risk

  Built and ran the eval pipeline for a payments-fraud LLM feature serving 40M requests/day. Mentors on retrieval and eval calibration.

• Dr. Marcus Hoel

  Principal ML Engineer · ex-Anthropic Applied

  Shipped tool-use agents into a production support workflow at a F500 insurer. Mentors on agent loops and budget governance.

• Priya Nair

  Platform Lead · ex-Datadog

  Owned the model-serving control plane for an observability product. Mentors on deployment, rollback, and cost accounting.

• Joaquín Mendes

  Founding Engineer · independent

  Author of the open-source eval harness used in the cohort capstone. Mentors on CI integration and regression detection.

Each mentor commits to a maximum of six learners per cohort. Mentor calibration data is shared with mentors directly; the aggregate distribution is on the transparency report.

Tuition

Regional pricing — tuition is set by the region of residence, not by employer. The outcome credit column is the percentage of tuition refunded if you do not place against the stated outcome inside the published 90-day window.

Questions engineers usually ask

What does the entrance assessment actually test?

Three things: a 60-minute Python + HTTP coding exercise, a short written reasoning task on a system trade-off, and a 25-minute conversation with a mentor. The rubric is published; we do not test trivia.

Can I take this part-time while employed?

Yes. The cohort is built for working engineers — 12–15 hrs/wk, two live sessions per week, asynchronous reviews. Fewer than 8% of past learners reduced employer hours during the cohort.

What happens if I do not finish the capstone in 14 weeks?

You may defer once to the next cohort at no additional tuition. A second deferral converts the seat to audit access. The full policy is in the enrolment agreement.

Is the credential recognised by employers?

The credential is a W3C Verifiable Credential with an Open Badges 3.0 assertion and is signed by the issuer key listed in our JWKS. Employers verify it from a public URL — no portal account required.

What is the outcome credit clause?

If you do not place against the stated outcome band within the published 90-day window, the percentage of tuition listed under the outcome credit column on the tuition table is refunded automatically.