A privacy-first, local AI assistant for symptom triage, clinical image review, and medical document interpretation. Runs with Ollama models and a Streamlit interface.
Try a hosted demo (when available): https://your-deployed-url.streamlit.app
Medo-AI is built to give rapid, structured medical guidance locally on your machine. It combines a Streamlit front end with Ollama-powered local models to provide:
- Conversational symptom assessment and first-aid guidance.
- OCR extraction and interpretation of lab reports, prescriptions, and PDFs.
- Clinical image triage (wounds, rashes, burns) with urgency indicators.
- Local-only inference to keep patient data private (no default external API calls).
The repository contains the Streamlit app (app.py), model/config settings (config.py), and modular services under services/.
- Conversational assistant with structured responses: likely causes, immediate steps, and escalation criteria.
- Document intelligence: OCR, abnormal-value highlighting, and plain-language summaries.
- Image triage: urgency scoring, care recommendations, and vision-model support for richer output.
- Safety guard: automatic red-flag detection and escalation suggestions.
- Privacy-first: default on-device processing using Ollama.
Prerequisites
- Python 3.8+
- Ollama installed and on PATH
- 8GB RAM minimum (16GB recommended for vision models)
- Optional: Tesseract OCR for improved document parsing
- Install Ollama
Windows
# via winget
winget install Ollama.Ollama
# or grab installer from https://ollama.com/download
macOS
brew install ollama
# or download from https://ollama.com/download
Linux
curl -fsSL https://ollama.com/install.sh | sh
Restart your terminal after installation so ollama is available in PATH.
- Pull recommended models (adjust for your hardware)
ollama pull gemma2:2b # Recommended primary model (fast & compact)
ollama pull qwen2:1.5b # Lightweight fallback
ollama pull llava:7b # Optional: vision-capable model for image analysis
- Clone repo & install Python deps
git clone https://github.com/Pratham-r05/Medo-AI.git
cd Medo-AI
pip install -r requirements.txt
- (Optional) Install Tesseract OCR
Windows: download UB Mannheim build and add to PATH. macOS:
brew install tesseract
Linux:
sudo apt-get install tesseract-ocr
- Run the application
streamlit run app.py
Open http://localhost:8501 (Streamlit may choose another free port if 8501 is busy).
- Edit config.py to change default model names, thresholds, or prompt templates.
- The app will try the primary model and fall back automatically if not available.
- Enable vision features only if you have a vision-capable model and sufficient memory.
- app.py — Streamlit UI and flow control
- config.py — model names, prompt text, and thresholds
- requirements.txt — Python dependencies
- services/
- chat_service.py — LLM chat handling and formatting
- document_service.py — OCR and PDF parsing utilities
- vision_service.py — image preprocessing and vision model integration
- safety_guard.py — red-flag detection and scope checks
- f650c04eef1b4bf428e9.png — demo screenshot used in README
- "ollama: command not found": ensure Ollama is installed and PATH updated; restart terminal.
- "Model not listed": run
ollama listand re-pull models withollama pull <model>. - Streamlit port conflict: run with
streamlit run app.py --server.port <PORT>. - Poor OCR results: confirm Tesseract is installed and language packs are present.
- Medo-AI is an assistive tool and not a replacement for clinical judgement.
- Always escalate suspected life-threatening conditions to emergency services.
- The default configuration prefers local processing; if you add integrations that transmit data off-device, review privacy implications carefully.
Important: This software is for informational use only and is not a medical device. Consult licensed healthcare professionals for diagnosis and treatment.
Contributions are welcome. Ways to help:
- Improve medical prompts and accuracy
- Add localization and language support
- Enhance vision pipelines and labeling
- Add tests, CI, and packaging
Open issues or PRs on GitHub; refer to CONTRIBUTING.md if present.
See requirements.txt for pinned versions. Typical packages include streamlit, opencv-python, Pillow, pytesseract, PyPDF2, pandas, and numpy.
- Report bugs or request features via GitHub Issues.
- Streamlit community forum for deployment questions.
- Check IMPROVEMENTS.md for ideas and roadmap notes.
MIT — see the LICENSE file for details.
Built with a focus on local-first, privacy-preserving medical AI — powered by Streamlit & Ollama.