README.md

---
title: Medical Conversation Viewer
emoji: "🩺"
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.12.0
app_file: app.py
pinned: false
license: mit
datasets:
  - Mediform/seed_data_v5
---

# Medical Conversation Dataset Viewer

Interactive viewer for the **Mediform/seed_data_v5** dataset containing synthetic German doctor-patient conversations for medical ASR training.

## Features

- **Conversation Selection**: Browse through different medical scenarios
- **Variant Support**: View conversations in three formats:
  - `natural`: Natural dialogue flow
  - `inline_dictation`: Dialogue with inline doctor dictation
  - `post_dictation`: Dialogue with post-turn dictation
- **Step-by-Step Navigation**: Walk through conversations turn by turn
- **EHR Reference Tracking**: Watch Electronic Health Record categories populate as the conversation progresses:
  - **History (Anamnese)**: Patient history and symptoms
  - **Findings (Befunde)**: Examination findings and test results
  - **Treatment (Therapie)**: Treatment decisions and medications
  - **Plan (Prozedere)**: Follow-up plans and diagnostics
  - **Orders (Anordnungen)**: Lab orders, appointments, prescriptions

## How It Works

The conversations contain `<ref>` tags that link spoken content to structured EHR entries. As you navigate forward through the conversation, referenced items are added to their respective categories. Navigating backward removes them, showing how the medical record builds up during the consultation.

## Dataset

This viewer displays data from [Mediform/seed_data_v5](https://huggingface.co/datasets/Mediform/seed_data_v5), which contains:
- Synthetic German medical dialogues
- Multiple conversation variants
- Structured EHR annotations
- Medical terminology (boost terms)

## Local Development

```bash
pip install -r requirements.txt
python app.py
```

## License

MIT