27 EU countries · 23-column extraction · every row traceable
London Health-Economics Consultancy: AI-Assisted SLR and JCA Extraction
Two Claude pipelines for a London health-economics consultancy: SLR screening with a quoted rationale for every criterion, and JCA clinical-environment extraction across 27 EU countries into a 23-column, fully traceable schema.
- SLR · screenPer-criterion decision + quoted rationale
- SLR · groundEvery quote checked against its source
- JCA · 27 countriesGuidelines · HTA decisions · EMA product information
- JCA · 23 columnsEvery row traceable to country, document, page and quote
The problem
SLR screening and JCA evidence extraction are high-volume, repetitive and auditable. Reviewers read thousands of abstracts and full texts, every decision needs a reason, and a JCA dossier needs per-country population and comparator evidence from guidelines, HTA decisions and product information, with verbatim quotes and traceability.
What we built
An SLR screening pipeline that sends each study to Claude with the review’s eligibility criteria and returns per-criterion decisions with quoted rationales in the existing Excel layout, with batch mode, resume and grounding checks. A separate JCA pipeline runs in a Claude Code session with no API key, extracts population and comparator evidence for 27 EU countries from national guidelines, HTA decisions and EMA product information, and writes a 23-column schema where every row is traceable to country, document, page and quote.
Before
- Thousands of records to screen
- Every decision needs a cited reason
- 27 countries, three evidence streams
How it works
Ingest the library. PDFs, registry exports, EndNote files and saved search pages load by content, not by filename.
Screen against the criteria. Each study returns a per-criterion decision and a quoted rationale, in the review’s existing Excel columns.
Ground every quote. A post-run check verifies that each rationale quotes its source record.
Batch, resume, never pay twice. The whole library runs at once, at roughly half the API cost, and a crash resumes from the checkpoint.
Extract JCA evidence. Population and comparator evidence for 27 EU countries, from national guidelines, HTA decisions and EMA product information.
Validate and hand over. The output is checked against the published JCA report; sources that cannot be fetched go to a human upload queue.
After
27 EU countries · 23-column extraction · every row traceable
Stack
Systematic Literature ReviewJCAHealth EconomicsEvidence ExtractionClaudeEU HTALife SciencesUKWhy it holds up
Every decision is auditable: a per-criterion rationale with a quote, a row traceable to country, document, page and quote, and a validation run against the published JCA report. The model reads and extracts; it does not invent. Sources it cannot fetch go to a human queue.
Team adoption
- Who uses it now
- The health-economics research team running the SLR and JCA work.
- What we handed over
- Two pipelines, the review prompts, the 23-column JCA workbook, the validation run and the human upload queue.
- What changed
- From manual screening and per-country extraction to per-criterion decisions with quoted rationales and a traceable JCA sheet.
Related service: AI adoption for research teams
More case studies
Big 4 client · fully on-premise
Enterprise RAG Chatbot for a Big 4 Accounting Firm
Multimodal retrieval over a Big 4 firm’s full compliance library, tuned for citation instead of summary — every answer arrives with its source paragraph attached.
- Juniors reading thousands of pages per question
- Seniors acting as search engines
200k calls/month · 67% booking rate vs 77% human
24/7 AI Agent Call Center: 200k Calls a Month Across 60 US HVAC Brands
Voice agents orchestrating 200,000 calls a month across 60 US HVAC brands for a 5,000-person enterprise, booking 67% of calls against 77% for human agents — within 10 points of a trained team, around the clock. Built as a product, an AI front desk that answers, books and routes, not a pilot.
- Night and weekend calls unanswered
- Staffing every shift hard to justify
US$1M collected in the first 3 months
AI Payment Collection and Reminders
The companion product to the AI front desk: it chases unpaid invoices with polite, timed reminders and a payment link, and stops the moment the customer pays or replies. It collected over US$1 million for the client in its first three months.
- Chasing payment is awkward and late
- Billable time lost to collections
First-Sprint Data Scientist
A vetted engineer embedded in your team, a shipped deliverable by end of sprint 1.
Pick a time below and you will speak with me directly. Bring the use case, the data and the constraints, and I will give you a straight view on feasibility, effort and the first thing to ship.
Book a callCat Yung
Hong Kong & London time zones
Book a call · 30 min with Cat Yung