DS Med Helper

A Streamlit-based web UI for physician medical documentation assistance using OpenAI API compatible LLM and ASR endpoints.

Features

• Scribe Mode: Record patient encounters or dictations and generate clinical notes automatically
• Note Edit Mode: Edit physician notes with AI assistance
• Synthesize Mode: Combine information from multiple sources (H&P, Consults, Studies, Progress Notes) to write comprehensive notes
• Session Persistence: Sessions are saved automatically and can be restored
• Configurable Endpoints: Connect to your custom OpenAI API compatible LLM and ASR endpoints
• Template System: Customizable note templates (H&P, Progress Note, Consultation, Discharge Summary)
• Sampling Controls: Adjust temperature, top_k, top_p, and min_p for LLM output
• Editable System Prompt: Customize LLM behavior through the Settings UI

Requirements

• Python 3.9+
• Streamlit
• PyYAML
• aiohttp

Installation

# Create virtual environment (recommended)
python -m venv venv
venv\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements.txt

Configuration

Edit config.yaml to configure your endpoints and LLM parameters:

server:
  host: "0.0.0.0"  # Listen on all interfaces
  port: 8501

llm:
  endpoint: "http://localhost:8080"
  api_key: ""  # Optional: Bearer token for authenticated endpoints
  model: "google/medgemma-27b-text-it"
  system_prompt: "You are a medical documentation assistant..."
  max_tokens: -1         # -1 for unlimited
  temperature: 0.8
  top_k: 40
  top_p: 0.95
  min_p: 0.05

stt:
  endpoint: "http://localhost:8000"
  api_key: ""  # Optional: Bearer token for authenticated endpoints
  model: "google/medasr"

session:
  max_history: 100
  storage_file: "sessions/session_data.json"

Runtime Settings

Many settings can be adjusted through the Settings UI:

• Endpoints (LLM and STT)
• System prompt for the LLM
• Model names
• Sampling parameters (temperature, top_k, top_p, min_p)
• Session management

Running

# Run the app
streamlit run app.py

# Or with custom host/port
streamlit run app.py --server.host 0.0.0.0 --server.port 8501

Templates

Note templates are stored as .txt files in the templates/ folder. Edit these files to customize the system prompts for each note type.

Usage

Scribe Mode

1. Select a note template from the dropdown
2. Record audio using the browser's audio input OR upload a file
3. Click "Transcribe Audio" to convert speech to text
4. Review/edit the transcript if needed
5. Optionally add context/instructions
6. Click "Generate Note" to create the clinical note
7. Download the recording if needed

Note Edit Mode

1. Paste your original clinical note in the left panel
2. Optionally select a template for context
3. Enter revision instructions (e.g., "Summarize to 3 bullet points")
4. Click "Generate Edited Note"

Synthesize Mode

1. Enter synthesis instructions (optional)
2. Fill in source information fields:
   • History and Physical
   • Consult Note(s)
   • Studies and Procedures
   • Progress Note(s)
3. Select a note template
4. Click "Generate Synthesized Note" to create a comprehensive clinical note from multiple sources

Settings

• Configure endpoints for LLM and STT servers
• Edit system prompt for LLM behavior
• Adjust sampling parameters (temperature, top_k, top_p, min_p)
• View, create, or clear sessions

Session Management

• Each browser maintains its own selected session via URL parameters (?session_id=xxx)
• Use the session picker dropdown at the top to switch between sessions
• Bookmark the URL to return to your active session after closing the browser
• "New Session" button creates a fresh session for the current browser
• Session history shows all sessions with ability to switch or delete

Access

When running on your home server:

• Local: http://localhost:8501
• Remote (VPN): http://<server-ip>:8501