Project 42Start learning
Module 16 of 16 · 75 min

AI Foundations Capstone

Plan, execute, verify, and explain a complete AI-assisted workflow with evidence that another person can review.

Core concept

By the end

You will be able to

  • Select a useful, bounded objective and identify data and authority limits.
  • Create a reproducible prompt or workflow plan with acceptance criteria.
  • Collect source, tool, output, and verification evidence.
  • Recover from a meaningful failure or document a tested recovery path.
  • Present the result and reflect on correctness, evidence, safety, reproducibility, and communication.
01

Choose a bounded objective

Choose a real but low-risk task that produces a reviewable result. State the user, decision, deliverable, acceptance criteria, deadline, and what is out of scope.

Classify the data before using an AI system. Remove secrets and unnecessary personal information, identify authoritative sources, and list any action that requires human approval.

Objective brief
text
User and need: [WHO / WHY]
Deliverable: [REVIEWABLE OUTPUT]
Acceptance criteria: [TESTS]
Data allowed: [CLASSIFICATION / REDACTIONS]
Authority: [READ / WRITE / APPROVAL]
Out of scope: [BOUNDARIES]
02

Build the prompt and workflow plan

Write a prompt or multi-step plan that separates instructions, trusted evidence, untrusted input, constraints, output contract, and verification. Select the provider or tool surface from the work rather than habit.

Add checkpoints before expensive, external, destructive, or permission-changing actions. Define what the workflow should do when evidence conflicts, a tool fails, or acceptance criteria are not met.

03

Execute and preserve evidence

Run the workflow in small observable steps. Save the exact prompt or plan, source references, material tool requests and results, output versions, and timestamps needed for another person to reproduce the work.

Do not preserve secrets, hidden reasoning, or unnecessary personal data. Evidence should show inputs, decisions, outputs, and checks without expanding the privacy footprint.

04

Verify and exercise recovery

Check the deliverable against every acceptance criterion. Use authoritative sources, deterministic tests, calculations, or direct inspection of the target system rather than asking the same model whether it succeeded.

Record one failure encountered and corrected, or safely simulate a likely failure. Show the stop condition, diagnosis, correction, rerun, rollback, or escalation and the evidence that the final state is acceptable.

05

Present, hand off, and reflect

Package the objective, workflow, evidence, deliverable, verification, recovery record, and limitations so a reviewer can trace each claim. Clearly separate facts, model-generated suggestions, unresolved uncertainty, and human decisions.

Score the work against the rubric, explain each score, and identify the most valuable improvement for a second version. Passing requires both the knowledge check and a capstone score of at least eighty percent.

Practice activity

Complete an evidence-based AI workflow

  1. Choose a low-risk research, writing, coding, analysis, or tool-use task and complete the objective brief.
  2. Create and execute a prompt or workflow plan with source, data, authority, checkpoint, and recovery boundaries.
  3. Preserve the required artifacts without secrets or unnecessary personal information.
  4. Verify every acceptance criterion independently and record a real or safely simulated failure-and-recovery cycle.
  5. Prepare a concise handoff, score each rubric criterion with evidence, and write the reflection.

What to produce

  • Objective and scope brief.
  • Prompt or workflow plan.
  • Source, tool, and result evidence.
  • Verification and recovery record.
  • Reflection and reviewer-ready handoff.

Reflect before continuing

What changed between your initial plan and the verified result, and what evidence justified that change?

Applied capstone

Evidence-based practical AI workflow

Pass at 80%

Demonstrate a complete, safe, reproducible AI-assisted workflow and submit traceable evidence for review.

Required artifacts

Provide a path, URL, or short stable reference for every artifact.

Evidence rubric

Award whole points from 0 to the criterion maximum. The total is checked against the published rubric.

Correctness

The deliverable satisfies the stated objective and acceptance criteria, with limitations identified.

Evidence required
  • Acceptance-criteria results
  • Final deliverable or stable reference

Evidence

Material claims and decisions are traceable to authoritative sources, tool results, or direct observations.

Evidence required
  • Source references
  • Relevant execution and verification records

Safety

Data, permissions, approvals, and consequential actions stay inside explicit boundaries.

Evidence required
  • Data classification and redactions
  • Authority and approval boundaries

Reproducibility

Another person can repeat the workflow and understand how failure was detected and recovered.

Evidence required
  • Prompt or workflow plan
  • Verification and recovery record

Communication

The handoff clearly separates facts, generated suggestions, decisions, uncertainty, and next steps.

Evidence required
  • Reviewer-ready handoff
  • Reflection and improvement

Evidence

Sources and verification

Knowledge check

Make it stick.

Pass at 80%

Choose the strongest answer for each question. Your attempts become part of your device-local transcript.

01What makes a task suitable for this capstone?
02Which evidence best supports reproducibility?
03How should a capstone verify an AI-generated factual claim?
04What must be included when no real failure occurs?
05When is the capstone module complete?