FAST-06 Projects: Occupation-Specific AI Role Labs
Provider: 7formations | Course: FAST-06 | Instructor: Wadood Chaudhary | One capstone and two mini projects
Projects extend the role playbook into daily practice. Scoring uses Not yet / Meets / Strong. Completion requires Meets on every capstone criterion; mini projects are formative unless the employer asks for scoring. One resubmission allowed. Fictional or sanitized material only; never applicant or employee records, client financial data, PHI, or customer personal information (G.L. c. 93H, 201 CMR 17.00).
Capstone: My Role Playbook in Use (21 days)
Brief. Put your ten-procedure playbook to work for 21 days. Use at least six procedures on real tasks (with sanitized inputs), log each use, improve procedures based on what you learn, and publish a team-ready version a same-occupation colleague has followed.
Deliverables
- Playbook v2: ten procedure cards (W2 format) with changes marked.
- Use log: 20+ uses across 6+ procedures (date, procedure, time by hand estimate, time with AI incl. check, issues caught).
- Data-safety log: for each use, what was stripped (category only, never the data).
- Two colleague tests: a same-occupation colleague followed two procedures cold; their notes.
- Retire list: procedures dropped or marked "no AI" and why.
- One-page summary for your manager: time saved, quality issues, recommended team adoption.
Step plan
| Step | When | What you do |
|---|---|---|
| 1 | Days 1-7 | Use six procedures; log every use |
| 2 | Days 8-14 | Fix vague checks and steps; colleague test 1 |
| 3 | Days 15-21 | Continue use; colleague test 2; retire list |
| 4 | Day 21 | Playbook v2, summary, submit |
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Role relevance | Generic tasks | Ten procedures match real tasks in the job family | Colleagues adopt at least two |
| Verification | Checks vague | Every card has a concrete check; issues logged | Checks caught real errors, with prompt fixes recorded |
| Data safety | Sensitive data used | Role-specific data excluded; strip log complete | Placeholders and strip lists built into prompts |
| Usability | No colleague tests | Two cold tests passed | Colleague followed with no questions |
| Measured value | No log or no check time | 20+ logged uses with check time | Honest retire list; team recommendation with numbers |
Strong vs weak example (one card, bookkeeping)
| Weak | Strong | |
|---|---|---|
| Task | "Client emails" | "Monthly variance explanation for small-business clients" |
| Data rule | "Be careful" | "Strip client and business names, account digits, SSN/EIN; use category totals and my notes only" |
| Prompt | "Explain this" | "Explain the top 3 variances in plain language using only my notes for causes; amounts exactly as given; 120 words" |
| Check | "Review it" | "Every amount matches the variance report; every cause appears in my notes; if not, delete the sentence" |
| Time | "Faster" | "By hand 15 min; AI 2 + check 4 = 6 min; 9 uses; 2 invented causes caught and removed" |
Mini project 1: Sanitization Kit for My Role
Brief. Build a one-page sanitization kit for your occupation: what to strip, what to keep, placeholders, and five before/after examples (fictional).
Deliverables. Strip list by category; keep list; placeholder set (e.g., [PATIENT], [CLIENT], [ORDER #]); five fictional before/after pairs; a "when not to use AI at all" list.
Step plan. (1) List the sensitive data in your role; (2) map each to strip or keep; (3) design placeholders; (4) write examples; (5) peer check by a same-occupation colleague.
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Completeness | Categories missing | All role-specific sensitive categories covered | Includes indirect identifiers (dates plus first name, etc.) |
| Practicality | Hard to use | One page, usable at the desk | Adopted by the team or employer |
| Accuracy | Examples leak data | Five clean examples | Examples include tricky edge cases |
Strong vs weak example. Weak: "Remove personal info." Strong: "Front office: strip names, DOB, phone, address, insurance IDs, appointment date with any name, tooth or treatment details tied to a person. Keep generic procedure names in templates only. Placeholders: [PATIENT], [DATE], [PHONE], [PROCEDURE]. Never use AI for: voicemail transcripts, referral letters, chart notes."
Mini project 2: Check-Step Upgrade
Brief. Take five procedures (yours or a colleague's) with weak checks and upgrade each to a concrete, testable check with an escalation rule. Prove each upgraded check catches a planted error.
Deliverables. Five before/after checks; for each, a fictional output with one planted error and evidence the new check catches it; escalation rule ("if X fails, do Y").
Step plan. (1) Collect weak checks; (2) name source and comparison; (3) plant errors; (4) test; (5) write escalation rules.
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Concreteness | Checks still vague | Each names a source and comparison | Each is sample-based or exhaustive as appropriate |
| Evidence | No test | Planted error caught for all five | Also tested on a real (sanitized) output |
| Escalation | None | Rule for each check | Rule includes who to tell |
Strong vs weak example. Weak: "Make sure the status report is right." Strong: "Compare each line's status, date, and owner with the meeting notes. Any date not in the notes becomes TBD; any at-risk item shown green triggers a call with the item owner before sending. Test: planted 'PPAP expected Nov 10' was caught."