AI / Prompt Support Engineer

Real portfolio pieces built during the course, not a certificate for its own sake. Each one is work you can show.

• LLM feature configuration recordOne page covering every setting your feature runs on, with the reason and the evidence behind each value, plus a decision log where each entry records the options considered, what was rejected and how you would know the decision was wrong.
• Prompt evaluation set with pass and fail criteriaA set of real cases with the context they should be answered from and explicit pass and fail criteria, split into deterministic code checks and human checklist items, reported by group with before and after results across repeated runs.
• API integration with error handlingWorking, provider-neutral integration code that classifies errors before reacting, backs off with jitter, caps total time, validates model output against facts you already hold, rejects citations to material that was never retrieved, and fails honestly with a defined fallback.
• Failure triage logOne row per real failure with the exact output, whether it reproduced and at what rate, a forced choice of cause class, the evidence behind that classification, whether it reached a customer, and the evaluation case you added as a result.
• Rollout and monitoring planA one page plan written before deploying: staged rollout with dwell times, what you watch at each stage, numeric rollback conditions decided in advance, the rollback mechanism, who is watching and when, and the removal condition for anything temporary.
• Draft edit analysis and baseline edit rateEvery change an experienced agent makes to an assistant draft, recorded with her own reason and a forced choice of where it would have to be fixed: the output contract, the code, or the interface. Carries the baseline edit rate by category, measured from the drafts and sent versions already stored, before anything was changed.