Harness
Prompt test suite
Frozen cases, pass/fail, and a diff when you change the wording.
Prompts that survive users: structure, tests, adversarial cases, and a harness so a wording change cannot silently regress.
Created by Baljeet Dogra
Harness
Frozen cases, pass/fail, and a diff when you change the wording.
Library
Three production tasks with schemas, few-shots and a changelog.
Expand a part for the syllabus. Content stays searchable when closed.
Roles, delimiters, task decomposition. Why notebook prompts fail on real users.
Selecting examples. JSON schemas, constrained decoding, repair loops.
Injection, jailbreaks you will actually see, and a frozen eval set.
Prompts in git. CI on the harness. Capstone: three tasks gated by tests.
You can get a decent answer in a playground. You cannot defend a change.
You own a feature whose quality is a prompt you have never tested.
Related: LLM API Integration · Applied GenAI Engineering
No. This is the prompting craft. Retrieval, agents and serving live in the longer production courses.
No. API credits are enough.
Four weeks to prompts you can test and roll back. Create an account to enrol.
Enrol now