A service that reads a PDF with blood test results, flags values outside the normal range, and gives recommendations. The report can be sent to a doctor.

Challenge

Turn a lab form into a clear report. Clients who can't send medical data to an external cloud need a version that runs on their own server.

What we did

  1. PDF recognition

    The lab form is converted into text and structured values.

  2. Value assessment

    Each value is marked "normal" or "out of range", with a confidence level.

  3. AI assistant

    It answers questions about the results through the ChatGPT API.

  4. Recommendations

    Advice is based on the abnormal values found.

  5. Sending to a doctor

    The full report goes to a specialist.

  6. Local model

    An open medical model based on LLaMA runs on the client's server, with no external APIs.

Stack

Next.js, Node.js, MongoDB, Docker, OpenAI API, LLaMA.

Result

The service is in development.