Audio-to-ICD (A2ICD)

➡️ Checkout running model on HuggingFace 🤗: NabJab/A2ICD.

The app uses whisper-1 (OpenAI) and gpt-4.1 to transcribe audio to text. To predict ICD-10 codes gpt-4.1 is used with a database lookup.

Input panels The Audio input panel allows users to record audio and transcribe it, while the Text input panel allows users to enter text directly.

Output panel The Output panel displays the predicted ICD-10 codes and links them to the ICD website. The codes are ordered by importance.

How to run

1. Setup API key(s)

Create a .env file in the root directory with the following content: OPENAI_API_KEY="your_openai_api_key"

2. Run app

• Option 1: Using uv (recommended): uv run mvp.py
• Option 2: Using python
  • Install dependencies: pip install -r requirements.txt
  • Run: python mvp.py

3. Open in browser Go to http://localhost:7860