AI-assisted medical document processing
Pulls the details out of incoming medical documents, so people check them instead of retyping them.
- For
- A healthcare organization
- My part
- Designer and developer
- Status
- Completed
The problem
A team was spending its days reading medical documents and typing the same details into other systems. It’s careful, repetitive work where a tired typo actually matters.
How it works, in 5 parts
A document arrives
Scanned and digital medical documents come in through the channel the team already uses.
Read the page
OCR turns each page into text, messy scans included.
Pull out the fields
AI extracts the specific details the team needs, as structured data rather than a paragraph.
A person checks
Someone reviews the extracted details next to the original before anything moves on.
File and route
The checked record continues into the team's workflow, and the document is filed where it belongs.
What I built
A document pipeline where AI does the first read. Each incoming document goes through OCR, then an extraction step pulls out the specific fields the team needs as structured data. A person reviews those fields next to the original document, fixes anything off, and approves it. Only then does the record move on.
The hard parts
Knowing where AI belongs. Extraction is a great job for AI. Deciding what’s correct is not. So the human review step isn’t a nice-to-have: it’s the design.
Real documents are messy. Skewed scans, handwriting, different layouts for the same form. The flow had to cope with partial results and make it obvious when something needed a closer look.
Privacy. Patient information stays inside the organisation’s own systems. Which is also why there’s no screenshot here.
Where it landed
Completed and in use. The team spends its time checking instead of typing, and documents move through noticeably faster.
What I learned
The best AI feature is often a boring one. Nobody needed a chatbot. They needed the typing to go away, with a person still in charge of what’s right.