AI for healthcare
Most healthcare data is text nobody has time to read: clinical notes, trial records, scanned documents. I build search and summarization on top of it so people get to the answer faster.
I usually build on LlamaIndex with Postgres and pgvector, and deploy on AWS, Azure or GCP. A person always stays in the loop. These tools draft and find things; they don't make the call.
Some of the work
Clinical documentation
Summarizing patient notes and ICD-10 codes into structured data, so clinical teams spend less time on manual entry.
Search over medical documents
RAG with LlamaIndex and pgvector, so people can get quick, sourced answers across trial documents and medical records.
From bookstore to hospital
The semantic search I first built for my own bookstore, Booksgrub, turned out to work just as well on healthcare documents.
Code you can look at
- Clinical summarization: unstructured notes into structured summaries
- medchunk: labels chunks of medical text, no LLM needed
- Clinical trials graph: trial data as a Neo4j knowledge graph
- Medical text on AWS: S3 events, Rekognition and Comprehend Medical