# AI project review worksheet

An original learning resource from Hashan Shalitha's Online Web Toolkit journal.
Use alongside Part 2: https://online-web-toolkit.com/blog/ai-singularity-future-software-engineers-students

This is a suggested review process, not an accreditation standard or hiring score.

## Project brief

- Intended user:
- Recurring problem:
- Existing non-AI approach:
- Permitted data sources and licenses:
- Consequences of an incorrect answer:
- Actions that require human review:
- What success would look like:

## Reproduction record

- Code revision and setup instructions:
- Model identifier, version, evaluation date:
- Prompt and dataset versions:
- Retrieval/document-order settings:
- Tool permissions and reasoning settings:
- Known costs and compute limits:
- A command or sequence that repeats one result:

## Evidence review

| Area                      | Evidence to attach                                                            | Pass / needs work / not assessed |
| ------------------------- | ----------------------------------------------------------------------------- | -------------------------------- |
| Problem and baseline      | User task and comparison with a simpler approach                              |                                  |
| Source support            | Answers linked to the correct source passages                                 |                                  |
| Evaluation                | Predefined acceptance rules, unfamiliar cases, and failure logs               |                                  |
| Boundaries                | Missing-information behavior, privacy limits, and untrusted-document handling |                                  |
| Reproducibility           | Versions, setup, and a repeatable example                                     |                                  |
| Engineering quality       | Tests, keyboard/accessibility checks, monitoring, and rollback where relevant |                                  |
| Independent understanding | An explanation and small modification completed without AI assistance         |                                  |

Do not turn these areas into an unexplained total score. A critical privacy or source-support failure cannot be canceled out by a polished interface.

## Five-minute project defense

1. Demonstrate a useful result and identify the evidence supporting it.
2. Demonstrate a failed or uncertain case and explain the limit.
3. Change a source or requirement and explain what should happen next.
4. Identify one AI-produced decision you accepted, changed, or rejected, and why.
5. Explain the most useful next experiment.

## AI assistance record

- Where AI was used:
- What I checked independently:
- What remains unverified:
- Feedback from another reviewer:
- Next revision:

Keep private student, customer, and employer information out of public portfolio artifacts.