Specialist assessment team · Document reviewThe problem
Before accepting a case, a specialist reviews a bundle of records that runs to thousands of pages. Fewer than 3% of those pages matter. They arrive in no order, mix GP, hospital and investigation records, and many older pages are handwritten. Finding the relevant pages, reconstructing the sequence of events and writing the assessment took about seven hours, half of it searching and sorting.
More viable cases went unreviewed each year than the team was able to review.
How it works
A pipeline takes the uploaded bundle and prepares it for the specialist. Pages are converted to text, with handwritten pages transcribed by two models and compared to flag low-confidence lines. Each page is classified by source and placed in date order. The system extracts conditions, dates, hospitals and treatments into a structured record, flags the significant pages and lets the specialist search the whole bundle in plain English.
- Handwriting transcribed by two models and cross-checked
- Pages classified by source and ordered into a chronology
- Conditions, dates and treatments extracted into a searchable record
- Significant pages flagged for review
- Plain-English search across the whole bundle
The results
- The specialist starts from the relevant pages in date order instead of a pile of thousands of pages.
- Handwritten notes are readable and flagged where the transcription is uncertain.
- Any question about the bundle can be asked in plain English and answered with page references.
- Review capacity rises fourfold at the full build, so viable cases stop going unreviewed.
