From QA, test or support engineering
QA and test engineers, SDETs, technical support and support engineers who already work close to customers and real failures.
What already counts
- Reproducing and debugging problems in customer environments
- Test thinking: edge cases, test sets and regression suites, which is most of what evals are
- Talking to frustrated customers and writing clear runbooks
- Reading logs and knowing how products break in the real world
What to add
- Building a system end to end, not only testing or fixing one
- Python fluency for production code, beyond test scripts
- Applied AI: retrieval, agents and guardrails
- Scoping and decomposition before any code exists
1Python and APIs as a builderWeeks 1 to 4
Write production-quality Python that reads messy data, calls APIs properly and answers questions with SQL.
Show for it: Three Code Lab exercises passed without hints and a small repo with tests and a README, written as if a customer's team will maintain it.
2Evals: testing for AIWeeks 5 to 7
Turn your testing strength into the most valued AI skill: golden sets, graders, regression suites and shadow-mode reports.
Show for it: The ticket triage take-home with an eval suite that runs in one command and a short note on which failures matter most to the customer.
3Build one AI system end to endWeeks 8 to 10
Own a whole system: retrieval or an agent, deployed in a container, with retries and logs you would want during an incident.
Show for it: A deployed search or assistant project with an eval score, a runbook and a two-minute demo video.
4From fixing to scopingWeeks 11 to 12
Move upstream: run discovery, find the real problem and write a scope before anything is built.
Show for it: A one-page scope from a Customer Room session, and a scorecard of solid or better on scoping.
5Reframe your experience and applyWeeks 13 to 14
Rewrite support and QA work in FDE terms and prepare for coding and behavioural rounds.
Show for it: A CV where your support and QA lines show integration debugging, customers served and tools you built, with numbers, plus four behavioural stories written down.