LLM107 · Evaluating, Testing and Guarding LLM Applications

Categories: AI & Automation
Wishlist Share

About Course

Measure the quality of LLM applications with real evaluation sets, automate regression tests, and add guardrails that protect users, data and the business.

Advance · 8 hours · 4 lessons, a lab, an assessed assignment and a portfolio piece

Who it is for

Developers and technical leads who are responsible for the quality and safety of LLM features before and after launch.

Prerequisites

LLM104 and LLM105; LLM106 is helpful for the agent-related examples.

What you will be able to do

  • Build an evaluation dataset with clear pass criteria for a specific LLM feature
  • Combine code-based checks and calibrated LLM-as-judge grading to score outputs
  • Automate prompt and model regression tests in a continuous integration pipeline
  • Implement input and output guardrails for personal data, off-topic requests and prompt injection, mapped to the OWASP Top 10 for LLM Applications

Lessons

  • Lesson 1 · Designing evaluations that matter
  • Lesson 2 · Grading and automated testing
  • Lesson 3 · Guardrails and defensive design
  • Lesson 4 · Lab and assignment

Lab: An evaluation suite and guardrails for a support assistant

You will create a 40-case evaluation set for a customer-support assistant, score it with deterministic checks and an LLM grader, and run it automatically on every change through GitHub Actions. You will then add guardrails that redact personal data, keep the assistant on topic and resist instructions embedded in user content, and verify them with tests.

Free tools: VS Code or Google Colab, promptfoo (open source, runs locally), pytest, Microsoft Presidio for PII detection, GitHub Actions free tier, and the Gemini API free tier or Ollama

Assessed assignment

Produce a complete quality and safety package for an LLM feature of your choice in a small business, such as a clinic appointment assistant or a store returns bot. Include a written quality bar with pass thresholds, an evaluation set of at least 40 cases, automated tests running in CI, at least three guardrails mapped to OWASP LLM risks, and a short incident playbook describing how the team detects, contains and fixes a harmful or wrong output after launch.

Criterion Weight What excellent work looks like
Evaluation design 30% Cases represent real usage, edge cases and safety risks; pass criteria are explicit and thresholds justified.
Grading quality 20% Deterministic checks are used wherever possible and the LLM grader is calibrated against human grades.
Automation 20% Tests run in CI on every change and block regressions with clear reports.
Guardrails and response 30% Guardrails are layered, tested and mapped to specific risks; the incident playbook is concrete and realistic.

Portfolio piece

A CI-integrated evaluation suite with tested guardrails and an incident playbook for a real LLM feature.

Self-paced, 12-month access. Questions are answered in writing through the student panel. A completion record is issued when you finish the course.

Next step

Continue with LLM108 · Deploying and Operating LLM Applications · 8 hours · $56.00

Continue to the next course

Prefer the whole program? Program in AI Application Development & LLM Engineering · $378.00 (10% less than buying the courses separately)

Show More

What Will You Learn?

  • Build an evaluation dataset with clear pass criteria for a specific LLM feature
  • Combine code-based checks and calibrated LLM-as-judge grading to score outputs
  • Automate prompt and model regression tests in a continuous integration pipeline
  • Implement input and output guardrails for personal data, off-topic requests and prompt injection, mapped to the OWASP Top 10 for LLM Applications

Course Content

LLM107 · Evaluating, Testing and Guarding LLM Applications (8 h)
Measure the quality of LLM applications with real evaluation sets, automate regression tests, and add guardrails that protect users, data and the business.

  • Lesson 1 · Designing evaluations that matter
  • Lesson 2 · Grading and automated testing
  • Lesson 3 · Guardrails and defensive design
  • Lesson 4 · Lab and assignment
  • LLM107 Assessment · Evaluating, Testing and Guarding LLM Applications
Scroll to Top