FAST-07 Projects: Industry-Specific AI Applications for Massachusetts Sectors
Provider: 7formations | Course: FAST-07 | Instructor: Wadood Chaudhary | One capstone and two mini projects
Projects turn the landscape sort, the sector labs, and the regulation-to-rule list into work products trainees use at their workplace. 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. Sanitized or fictional material only; never real customer, patient, or employee data (G.L. c. 93H, 201 CMR 17.00, HIPAA). Training, not legal advice.
Capstone: Industry Use-Case Brief and 45-Day Pilot Plan
Brief. Choose one AI application for a real, recurring task at your workplace. Write a one-page use-case brief (Lab 4 template), then extend it with a 45-day pilot plan your manager could approve. Pitch it to the cohort, revise from critique, and submit both.
Deliverables
- Landscape note (half page): three real AI applications in your industry, each classified mature, emerging, or unproven, with one line of evidence each.
- One-page use-case brief: task and current owner, AI method and approved tool, data rules, risks with checks and owners, measurable benefit.
- Pilot plan (one page): baseline measurement method; who pilots; weekly check routine; stop rule; review date and decision owner.
- Sample output: one AI-assisted output for the task made with fictional or sanitized input, plus the documented check (what you compared, what you fixed).
- Peer critique log: questions asked in your pitch and the changes you made.
Step plan
| Step | When | What you do |
|---|---|---|
| 1 | After session 1 | Pick the task; write the landscape note |
| 2 | After session 2 | Run the task once with fictional input; record the check |
| 3 | After session 3 | Add data rules from your W6 list; vendor questions if the AI is inside industry software |
| 4 | Session 4 | Pitch, critique, revise |
| 5 | Within 10 days | Pilot plan; submit |
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Industry accuracy | Uses or rules misdescribed | Uses classified with evidence; rules correctly named for the sector | Cites the specific rule section or guidance and the person who confirmed it |
| Task quality | Sample output fails the sector standard | Sample output meets the standard after a documented check | Check is built into the workflow (template, checklist) |
| Risk controls | Generic ("be careful") | Each named risk has a specific check and an owner | Includes a stop rule and an escalation path |
| Measurability | No baseline or indicator | Indicator, baseline, target, review date | Net benefit accounts for check time and rework |
| Data safety | Plan requires pasting restricted data | Approved tool; sanitized inputs; data rules applied | Shows how the rule is enforced (tool setting, template field) |
Strong vs weak example (Cape Harbor Inn, hospitality)
| Weak | Strong | |
|---|---|---|
| Task | "Reviews" | "Draft replies to online guest reviews (about 25 per week) for manager approval" |
| Risk | "AI might say something wrong" | "Reply offers an unapproved refund or comp. Control: manager approves every reply; prompt forbids offers" |
| Benefit | "Faster replies" | "Median time from review to reply: 52 hours now (4-week sample) to under 12, including approval" |
| Review | none | "Day 45, general manager decides continue, change, or stop" |
Mini project 1: Vendor AI Feature Evaluation
Brief. Pick one AI feature in software your industry uses (EHR ambient documentation, ERP or MES assistant, construction project assistant, claims system assistant, POS or reservation assistant, WMS or route optimization). Using only public vendor material and fictional screenshots, write a two-page evaluation with the questions you would put to the vendor in writing.
Deliverables. Feature description; landscape classification with evidence; data questions (what data, where stored, training use, retention, opt-out); testing questions (tested on businesses like ours, results); control questions (review, override, off switch, logging); your recommendation (pilot, wait, or decline) with reasons.
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Data questions | Missing or generic | Covers use, storage, training, retention, opt-out | Ties each question to a specific rule (93H, HIPAA BAA, NDA) |
| Evidence | Relies on marketing claims | Separates claims from evidence | Identifies what evidence would change the recommendation |
| Recommendation | No decision | Clear decision with reasons | Includes conditions for revisiting |
Example question (strong). "Does SmartReply use our guest messages to train or improve any model? If yes, can we opt out at the account level, and will you confirm that in our contract?"
Mini project 2: Sector Do/Don't Card and Five-Minute Team Briefing
Brief. Turn your W6 regulation-to-rule list into a one-page card for your team and a five-minute briefing script (as you would deliver at a shift start or toolbox talk). Get your compliance lead or manager to review it.
Deliverables. The card (8 to 12 testable lines, each with its source); the briefing script; two scenario questions to check understanding; reviewer's name and role and any changes they requested.
Rubric
| Criterion | Not yet | Meets | Strong |
|---|---|---|---|
| Testability | Vague lines | Every line checkable as done or not done | Each line has an example of compliance and a violation |
| Accuracy | Rules misstated | Sources correctly named | Reviewer confirmation documented |
| Usability | Too long or dense for a briefing | Fits one page; script under five minutes | Piloted with two coworkers and revised |
Strong vs weak line. Weak: "Be careful with patient info." Strong: "Never type a patient's name, date of birth, or chart number into any AI tool. Use the approved handout templates instead. (HIPAA Privacy Rule; practice policy 4.2)"