Lab AI

Lab AI was a research preview that took a lab report as a PDF and gave you a plain-language reading of it: each value, its unit, and where it sat against the reference range. We built it for patients trying to make sense of their own results, and for the care teams reading those results with them.

I built it after a run of frustrating telemedicine visits. The announcement on 1 May 2023 put our position plainly: “Will AI replace Doctors : YES / Will that happen today : NO”. The point was narrower than diagnosis. A report full of abbreviations and numbers can be made readable before anyone talks to a doctor. The demo video in that thread is the working proof; the SDK followed on 26 May 2023.

We never published usage numbers for the preview, and we are not going to make some up now.

How it worked

The pipeline had two halves. We parsed the PDF in the browser, on the client, through a usePDFParser hook. The extracted text then went to a hosted inference API at labai.tecmie.africa, which pulled out the values, units and ranges and returned them in a form we could render readably.

We packaged that client half as @tecmie/labdoc-sdk, “Labdoc SDK for React” on npm. It wraps a tRPC query client in React Context and exports LabProvider, usePDFParser and useLabInference, plus a Next.js higher-order component that wires up the query client and parser in one step. MIT licensed.

Two versions went out: 0.1.8 on 26 May 2023 and 0.1.9 three days later. The source, tecmie/labdoc-sdk, is 16 commits. Usage notes are on the Lab AI NodeJS SDK docs page, and the original technical proposal is in the resources below.