💬 When patients Google their symptoms or ask ChatGPT about their medications, they get answers — but not always the right ones.
Mana gives them the right ones. ✨
Every day, patients search the internet for answers about their health. They get blog posts, forum opinions, and AI-generated text that sounds confident — but isn't grounded in anything real. They leave more confused than when they started, or worse, make decisions based on information that was simply wrong.
Mana is different. It answers health questions directly from verified clinical guidelines — the same standards doctors actually rely on. Ask about your blood sugar target 🩸, your blood pressure medication 💊, or what a lab result means 🔬. Mana finds the answer in the actual guideline and gives it to you clearly, in your language.
🛡️ If the answer isn't in the guidelines, Mana says so. An honest "I don't know" is worth more than a confident wrong answer.
🙋 You ask a question
↓
🔍 Mana searches 2,600+ chunks from clinical guidelines
↓
📄 Finds the 6 most relevant passages via semantic search
↓
🤖 Answers ONLY from those passages — nothing else
↓
💬 Response in your language, grounded in the source
This is RAG (Retrieval-Augmented Generation) — the AI answers from the specific documents we've curated, not from general training data. Like an open-book exam where the book is the ADA Standards of Care 📚
| 📖 Guideline | 🏥 Organization | 📅 Year |
|---|---|---|
| Standards of Medical Care in Diabetes | ADA | 2026 |
| Hypertension Guidelines | AHA/ACC | 2025 |
| (actively expanding) 🚀 |
🔒 Currently in private beta
| 🔧 Layer | ⚙️ Technology |
|---|---|
| 🧠 LLM | GPT-4o-mini |
| 🗄️ Vector Database | ChromaDB |
| 🔢 Embeddings | OpenAI Ada-002 |
| 🔗 RAG Framework | LangChain |
| 🎨 UI | Gradio |
| 📄 PDF Processing | PyPDF |
| ☁️ Deployment | HuggingFace Spaces |
| 🐍 Language | Python 3.11 |
mana/
├── 📂 data/
│ ├── 📄 *.pdf # Clinical guidelines
│ └── 🗄️ chroma_db/ # Vector database (auto-generated)
├── 📂 src/
│ ├── __init__.py
│ └── 💬 prompt.py # Central prompt — single source of truth
├── 🚀 app.py # Main application
├── ⚙️ ingest.py # PDF processor & database builder
├── 📋 requirements.txt
└── 🔒 .env # API keys (never committed)
# 🐍 Create environment
conda create -n mana python=3.11.9 -y
conda activate mana
pip install -r requirements.txtAdd your OpenAI key to .env:
OPENAI_API_KEY=sk-...
Build the knowledge base 🧠:
python ingest.pyRun 🚀:
python app.py- 📥 Drop any guideline PDF into
data/ ▶️ Runpython ingest.py
The pipeline automatically:
- 🧹 Filters out reference lists and blank pages
- ✂️ Chunks text intelligently with overlap
- 🏷️ Tags each chunk with its source document
- 🔄 Rebuilds the entire database
No code changes ever needed. ✅
- 🛡️ Honesty over helpfulness — Mana will not answer outside its knowledge base. Ever.
- 🔍 Traceability — Every answer is grounded in a specific passage from a specific page.
- 🌍 Accessibility — Persian and English supported. Designed for patients, usable by clinicians.
- 🎯 Single source of truth — One prompt file, one database, one place to change anything.
- GOLD (COPD), GINA (Asthma), ESC Cardiac guidelines
- 👤 Patient health profile with consent-based data collection
- 👨⚕️ Doctor-facing clinical mode
- 📱 Telegram bot
- 📲 Mobile app
Mana provides information from clinical guidelines for educational purposes only. It does not provide medical diagnosis or personalized medical advice. Always consult a qualified healthcare professional for medical decisions.
© 2026 Mana Health — All rights reserved 💙