---
title: OTC MedGuide
emoji: 💊
colorFrom: blue
colorTo: green
sdk: streamlit
sdk_version: "1.28.0"
app_file: app.py
pinned: false
---

# OTC MedGuide

A Streamlit-based conversational assistant that provides informational guidance about over-the-counter (OTC) medications and dietary supplements using **Meditron-7B (epfl-llm/meditron-7b)**.

## Features


✅ **Meditron-First Reasoning** - Uses Meditron-7B for medical knowledge and safe recommendations
✅ **OTC Drug Database Enrichment** - Only uses dataset for brand names, dosage, warnings, interactions
✅ **Filtered Recommendations** - Avoids irrelevant drugs (e.g., topicals for headache)
✅ **Table Output** - All recommendations and comparisons are shown in a clear table
✅ **Full OTC Coverage** - Dosage, safety, mechanisms, comparisons, multi-turn follow-ups
✅ **Context Handling** - Tracks previous user input (symptoms, age, pregnancy, medications) for follow-ups
✅ **Friendly, ChatGPT-like UI** - Input at the bottom, single send, sidebar features (user guide, disclaimer, dataset view)

## What It Does


- Answers questions about OTC medications using Meditron's medical knowledge
- Recommends appropriate OTC drugs for common symptoms (never irrelevant drugs)
- Provides general information about medications (uses, warnings, dosage, interactions)
- Includes safety disclaimers for medical concerns

## What It Does NOT Do


- ❌ Diagnose medical conditions
- ❌ Prescribe medications
- ❌ Replace professional medical advice
- ❌ Recommend prescription drugs

## Requirements

- Python 3.9+
- CUDA-capable GPU (recommended for faster inference)
- At least 8GB RAM

## Installation

### 1. Clone or Download

```bash
cd chatbot
```


### 2. Create Virtual Environment

**Windows (PowerShell):**
```powershell
python -m venv .venv
.\.venv\Scripts\Activate.ps1
```

**Linux/Mac:**
```bash
python -m venv .venv
source .venv/bin/activate
```

### 3. Install Dependencies

```bash
pip install -r requirements.txt
```

**Note:** First-time installation may take 10-15 minutes as it downloads the Phi-3 model (~2.3GB).
**Note:** First-time installation may take 10-15 minutes depending on model downloads and your environment.

### Run Tests

After installing dependencies, run the unit tests to verify core intent and response behavior:

```bash
pytest -q
```

## License

This project uses open-source models and libraries. See individual component licenses.

## Support

For issues or questions:
- Check the Troubleshooting section
- Review application logs in the Streamlit UI
- Verify `otc_dataset.json` is valid JSON
- Ensure Python 3.9+ is installed

---

**Last Updated**: December 17, 2025
**Model**: epfl-llm/meditron-7b
docker run -p 8501:8501 otc-chatbot
```

Visit `http://localhost:8501` in your browser.

## Project Structure

```
chatbot/
├── app.py                    # Main Streamlit application
├── otc_dataset.json         # OTC drug database (100+ drugs)
├── requirements.txt         # Python dependencies
├── Dockerfile               # Docker configuration
├── README.md               # This file
└── .venv/                  # Virtual environment (created during setup)
```

## Example Interactions

### Example 1: Greeting
```
User: Hi
Bot: Hello! I'm an OTC drug recommendation assistant. You can ask me about symptoms, medications, or general health concerns.
```

### Example 2: Symptom Inquiry
```
User: I have a headache
Bot: For headaches, you might consider acetaminophen or ibuprofen. 
Acetaminophen (Tylenol) is effective for mild to moderate headaches and is generally well-tolerated.
Ibuprofen (Advil, Motrin) is an anti-inflammatory that works well for tension headaches...
[includes safety warnings]
```

### Example 3: Medication Information
```
User: What are the side effects of ibuprofen?
Bot: Ibuprofen warnings include:
- GI bleeding risk (especially with long-term use)
- Cardiovascular risk with chronic use
- Kidney effects in susceptible individuals
- Should be taken with food to minimize stomach upset
[includes appropriate disclaimers]
```

## Model Information

- **Model**: `epfl-llm/meditron-7b`
- **Size**: ~14GB (quantized)
- **Type**: Medical language model
- **Context Window**: 4096 tokens
- **Capabilities**: Medical reasoning, drug information, safety-aware responses

### Why Meditron?

- ✅ Specialized for medical Q&A and reasoning
- ✅ High-quality medical knowledge and safety awareness
- ✅ Excellent for OTC medication recommendations
- ✅ Local inference for privacy and reliability

## Configuration

Key parameters in `app.py`:

```python
MODEL_ID = "epfl-llm/meditron-7b"
MAX_NEW_TOKENS = 512              # Max tokens per response
TEMPERATURE = 0.6                 # Higher = more creative
TOP_P = 0.95                      # Nucleus sampling parameter
```

Adjust these values in the code if needed.

## Safety & Compliance

- **Medical Disclaimer**: Automatically displayed to users
- **Non-Diagnostic**: Does not diagnose or treat diseases
- **Informational Only**: Provides general guidance, not medical advice
- **Professional Referral**: Recommends seeing healthcare providers for serious symptoms
- **Pregnancy/Breastfeeding**: Includes safety warnings

## Performance Notes

- **First Run**: Model downloads (~14GB). Subsequent runs load from cache (much faster)
- **Inference Time**: 5-15 seconds per response (depends on hardware)
- **Memory**: ~8-16GB RAM during inference
- **GPU**: Highly recommended for <10s response times

## Troubleshooting

### Model Takes Too Long to Load
- First-time download is large. Check internet connection
- For CPU-only systems, expect longer load times
- Consider GPU access for faster inference

### Out of Memory Errors
- Reduce `MAX_NEW_TOKENS` to 128 in `app.py`
- Close other applications
- Use Docker for isolated environment

### Streamlit Port Already in Use
```bash
streamlit run app.py --server.port 8502
```

### Import Errors
```bash
pip install --upgrade transformers torch
```

## File Descriptions

### `app.py`
Main application entry point. Contains:
- Model loading and caching
- OTC dataset retrieval and matching
- Response generation with Meditron-7B
- Streamlit UI components
- Chat history management

### `otc_dataset.json`
Structured database of 100+ OTC drugs with fields:
- `generic_name`: Drug's generic name
- `brand_names`: Brand names (e.g., Tylenol, Advil)
- `category`: Drug category (Pain/Fever, Allergy, etc.)
- `common_use`: Primary uses
- `warnings`: Safety warnings
- `interactions`: Known drug interactions
- Other fields for dosage, contraindications, pregnancy flags, etc.

### `requirements.txt`
Python packages required:
- `streamlit`: Web UI framework
- `transformers`: HF transformer models and pipelines
- `torch`: Deep learning framework (with CUDA support)
- `pandas`: Data manipulation
- `accelerate`: Distributed computing for faster inference

### `Dockerfile`
Container configuration for deployment:
- Python 3.11-slim base
- Installs all dependencies
- Copies application files
- Exposes port 8501
- Runs Streamlit server

## License

This project uses open-source models and libraries. See individual component licenses.

## Support

For issues, questions, or improvements:
1. Check the Troubleshooting section
2. Review application logs in Streamlit UI
3. Verify `otc_dataset.json` is valid JSON
4. Check Python version compatibility (3.9+)

---

**Last Updated**: December 16, 2025
**Last Updated**: December 17, 2025
**Model**: epfl-llm/meditron-7b