Evaluating AI Models
Course 07 · Capstone Project

Build Your AI Evaluation Framework

Apply everything from the course to design and document a complete, deployable evaluation system for a real use case.

90 min · Synthesizes all 6 modules

What You're Building

Evaluation frameworks don't exist in the abstract. They're purpose-built for specific models, tasks, and organizational contexts. In this capstone, you'll choose a real AI use case and design a complete evaluation system: from pre-deployment benchmarks through continuous production monitoring, covering automated metrics, human review protocols, red-teaming plans, and drift detection.

The output is a documented evaluation plan you can actually use, adapt, or share with a team. It should be specific enough that someone who hasn't read this course could implement it.

Choose Your Scenario

Select one of the following scenarios, or bring your own real use case. The frameworks you'll build are structurally similar across domains, but the specific metrics, thresholds, and risks differ substantially.

Legal
Contract Review AI
An LLM that flags risk clauses and summarizes contract terms for legal teams.
Healthcare
Clinical Documentation
An AI that drafts clinical notes from physician dictation and structured data.
Customer Service
Support Chatbot
A conversational AI handling tier-1 customer queries for a financial services firm.
Engineering
Code Review AI
An AI that reviews pull requests for bugs, security issues, and style violations.

The Four-Phase Framework

Your evaluation framework should address all four phases. Work through each phase in order, since later phases depend on decisions made in earlier ones.

Phase 1
Define Quality Dimensions and Minimum Standards

List the 4-6 quality dimensions that matter most for your chosen use case. For each dimension, define what "good" means in plain language, and specify a minimum acceptable threshold that, if breached, would block deployment.

Deliverable: Quality matrix: dimension, definition, minimum threshold, and measurement method for each.
Phase 2
Design Pre-Deployment Evaluation

Select the benchmark datasets and automated metrics you'll use before deployment. Specify any domain-specific test sets you'd need to create. Define your human evaluation protocol: rubric, reviewer criteria, sample size, and inter-annotator agreement target.

Deliverable: Pre-deployment evaluation plan with specific metrics, tools, and go/no-go criteria.
Phase 3
Build a Red-Teaming Plan

Identify the top 5 adversarial scenarios specific to your use case. For each: describe the attack type, the potential harm if successful, and the specific test prompts or strategies you'd use. Define what a "pass" looks like for each scenario.

Deliverable: Red-teaming playbook with 5 attack scenarios, harm descriptions, and pass/fail criteria.
Phase 4
Design Continuous Monitoring

Specify your production logging strategy, automated scoring pipeline, human review sampling rate, metric aggregation cadence, and alert thresholds. Define what events trigger each alert level, and who owns the response.

Deliverable: Monitoring runbook with logging spec, metrics dashboard design, alert matrix, and escalation owners.
Framework Coverage Radar Interactive

Rate how thoroughly your framework addresses each evaluation dimension. Use this as a self-check before finalizing.

Coverage score: 0/100

Evaluation Plan Rubric

Use this rubric to self-assess your framework before finalizing it. A strong framework scores 70 or above. Anything below 50 has critical gaps that should be addressed before using the framework for real deployment decisions.

Self-Assessment Rubric
Quality Dimensions
Are your quality dimensions specific to the use case, not generic? Does each have a measurable definition, not just a label?
20 pts
Metric Selection
Do you use both automated and human metrics? Are the chosen metrics valid for the specific task type (classification vs. generation vs. RAG)?
20 pts
Red-Teaming Depth
Do your adversarial scenarios cover both direct attacks (jailbreaks) and indirect failures (edge cases, distribution shifts)? Are harms domain-specific?
20 pts
Continuous Monitoring
Is your monitoring plan operationally realistic? Does it specify data retention, sampling rates, alert ownership, and escalation paths?
20 pts
Feedback Loop
Does your framework describe what happens when evaluation signals failures? Is there a process for translating evaluation findings into model improvements?
20 pts
Total 100 pts

Common Framework Gaps to Avoid

Watch Out For
  • Generic metrics without thresholds: "we'll track accuracy" is not a plan. Specify the metric, the measurement method, and the minimum threshold.
  • Red-teaming as a checklist: five attack types with no pass/fail criteria isn't a test plan. Define what constitutes a failure for each scenario.
  • Monitoring without ownership: a dashboard that no one is responsible for watching doesn't create safety. Every alert level needs a named owner and a response SLA.
  • Evaluation that ends at launch: the most common gap. A framework that only covers pre-deployment gates will miss post-deployment drift and edge cases that only appear in real traffic.

Connecting to Real Research

Your framework should reflect the current state of the field. Several concepts from this course map directly to published research you can cite when presenting your framework to stakeholders.

Certificate of Completion
Your Name
Click to edit
has successfully completed
Evaluating AI Models
Why Evaluation Matters Benchmarks Human Evaluation Domain-Specific Red-Teaming Continuous Monitoring Capstone
Click any field on the certificate to edit