Domain packs
Run the course with data from your own industry
Every lesson uses one default running example: a support-ticket service with tickets, customer accounts, messages and resolutions. It’s simple, everyone understands it, and it needs no regulated data.
Learners from a specialised industry get more out of the course by swapping in their own domain. Each pack below shows:
- how your domain’s entities map onto the running example
- public datasets you can use in each phase
- the compliance topics for week 20
- capstone ideas
Rule for every pack: use public or synthetic data only. Never put real personal, patient or customer data into course projects, LLM prompts or public repos.
Pharma and life sciences
Entity mapping
| Running example | Pharma equivalent |
|---|---|
| ticket | adverse-event report, or a medical-information enquiry |
| account | product (drug), or a trial sponsor |
| status | case seriousness / outcome / processing status |
| plan changes (SCD2) | label versions, formulation or manufacturer changes |
| help-centre articles (RAG) | drug labels (SPL), prescribing information, SOPs |
Public data by phase
| Weeks | Use |
|---|---|
| 2–3 (CLI) | curl + jq against the openFDA APIs (api.fda.gov/drug/event.json, /drug/label.json) and the ClinicalTrials.gov API v2 |
| 5–8 (API) | Build a service over your own curated subset: products, events, enquiries |
| 12–15 (data) | FAERS quarterly extracts (adverse events) → medallion pipeline. Dimensions: drug, reaction (MedDRA preferred terms), reporter type, country. SCD2 on product labels using DailyMed label versions |
| 16–19 (AI) | RAG over openFDA or DailyMed drug labels with citations; structured extraction of drug, dose, reaction and seriousness from synthetic case narratives; an agent over ClinicalTrials.gov |
| 21–22 (stats) | Disproportionality analysis (PRR / ROR) on FAERS, and why reporting bias makes FAERS hypothesis-generating rather than proof (selection bias and confounding in practice) |
Week 20 compliance:
- GxP and computer system validation
- 21 CFR Part 11 and EU Annex 11 (audit trails, e-signatures)
- pharmacovigilance timelines
- GDPR for trial-participant data
- when an AI feature becomes “validated software”
Capstone ideas
- Drug-label comparison and Q&A tool with change tracking across label versions
- Adverse-event signal explorer: pipeline, PRR/ROR, LLM-drafted signal summaries with a human reviewer in the loop
- Clinical-trial landscape explorer: sponsor, site and indication analytics plus cited protocol Q&A
- Medical-information enquiry triage: classify, draft a response from approved content, escalate off-label questions
Healthcare
Entity mapping: ticket → encounter or referral · account → provider/practice · status → referral status · plan changes → provider network or system changes · help-centre articles → clinical guidelines or practice policies.
Public data
- Synthea: synthetic patients and EHR records (FHIR/CSV); use it for weeks 12–15
- CMS public datasets: provider directories, quality measures
- FHIR sandbox servers for API weeks
- MIMIC needs credentialed access; only use it if you have that access
Week 20 compliance: HIPAA (PHI, minimum necessary, BAAs, audit logs, breach notification), GDPR special-category data, and national health-data rules.
Capstone ideas: a referral-tracking co-pilot, appointment no-show prediction with outreach, or guideline Q&A with citations.
Fintech and payments
Entity mapping: ticket → transaction or dispute · account → customer account or merchant · status → dispute or fraud-review status · plan changes → risk tier or KYC level changes · help-centre articles → policy and regulatory docs.
Public data
- PaySim (synthetic mobile-money transactions)
- the public credit-card fraud dataset (Kaggle)
- SEC EDGAR filings for RAG
- open banking sandbox APIs
Week 20 compliance: PCI DSS scope reduction, KYC/AML data handling, SOX controls, and model risk management for AI decisions.
Capstone ideas: dispute triage with evidence summarisation, a fraud-alert review queue with LLM explanations, or a filings Q&A tool.
E-commerce and marketplaces
Entity mapping: ticket → order or return · account → customer or seller · status → order/return status · plan changes → seller tier changes · help-centre articles → returns and shipping policies.
Public data
- the Olist Brazilian e-commerce dataset (orders, reviews, sellers)
- Instacart basket data
- Amazon review datasets
Week 20 compliance: GDPR/consumer law, PCI scope when taking payments, accessibility law (European Accessibility Act).
Capstone ideas: a returns-reason analyser, a review-insight dashboard with LLM theme extraction, or a seller-support co-pilot.
Bring any other domain
Answer these five questions in week 3, then map your answers onto the running example:
- What’s the core event (your “ticket”)?
- What’s the main entity that changes over time (your “account” with SCD2)?
- What documents would an assistant need to answer questions from (your RAG corpus)?
- What regulation governs the data?
- Where can you get public or synthetic data?