AI APPLICATION ENGINEERING
AI Prompt & Context Engineering Institute
A support assistant is using the wrong return policy. Rewrite the prompt, choose the right evidence, and compare both versions on five identical tickets.
Start the support investigationNorthline return decisions
Northline’s current policy conflicts with an old FAQ and a customer’s claim. The two sides start with authored responses. Scoring those fixtures evaluates the recorded answers. Prompt edits leave the responses as they were until you replace them.
Expected output for this ticket
Evaluation report
Exact fields compare ticket, decision and days. Evidence checks require an applicable source and an exact excerpt in the current context packet. Human rubric scores are reported separately.
Structured output studio
Validate with JSON Schema 2020-12. The opening contract checks nested evidence objects, rejects extra fields, and accepts an integer or null for elapsed days.
Cost and latency assumptions
Enter rates from your provider or your own measurements. The token estimate uses roughly four characters per token and includes the assembled prompt for the selected ticket.
Practice library
Research notebook
Saved versions
Weekly research challenge
Learn prompt and context engineering
Design prompts, assemble sources, validate machine-readable output, and run regression checks on a shared dataset.
Browse interactive guides · AI learning path