MedGuide: AI clinical summaries from patient records
A Next.js application and document pipeline that turn patient records in any format or language into structured clinical summaries, using Gemini for extraction.
The product problem
Clinics, hospitals and healthcare teams receive patient records as PDFs, scans and photographs, in whatever language and format the originating hospital used. A clinician may have only a fifteen-minute appointment to understand a long and unfamiliar file. The product has to turn that material into a structured summary a clinician can read quickly, while respecting the handling constraints that medical data carries.
The delivered scope
The web application is built in Next.js and TypeScript with Supabase. It covers upload, a document viewer, structured report dashboards, translations, follow-up chat, PDF export and subscription billing. A NestJS backend runs the document pipeline as queued jobs: extraction, validation, translation with layout preservation and report building. Gemini on Google Cloud performs the extraction. The product is built around end-to-end encryption, zero retention and GDPR compliance.
Treating extraction as a pipeline, not a prompt
Medical documents vary in layout, quality and language, so a single model call is not enough. Separating upload, extraction, validation, translation and report building gives each step its own checks and failure handling. For a similar AI integration, scope should cover the documents the system will actually receive, how output is validated before a professional reads it, and how sensitive data is handled at every step.
What shipped
A clinical document intelligence product that turns mixed-format patient records into structured summaries through a Next.js application and a Gemini-backed processing pipeline.
Visit the project ↗