MedBot – AI-Powered Health Assistant & Pill Dispenser

MedBot is an intelligent healthcare assistant that combines AI-driven medical chatbot with a Raspberry Pi–controlled pill dispenser. It uses Google Vertex AI for generating responses and Google Cloud Speech-to-Text for voice input.

🏆 3rd Place Winner – Engineering Clinics Expo 2025 (500+ projects)

Features

• Voice Input – Converts speech to text using Google Cloud Speech API
• Chatbot Responses – Powered by Vertex AI generative models
• User Interface – Tkinter + ttkbootstrap with splash + chat window
• Chat History – Scrollable user–bot conversation logs
• Pill Dispenser Integration – Raspberry Pi hardware extension

Tech Stack

• Python 3.10+
• Tkinter + ttkbootstrap – UI framework
• Vertex AI – AI chatbot backend
• Google Cloud Speech-to-Text – Real-time speech recognition
• PyAudio – Microphone input
• Pillow (PIL) – Image handling in UI

Setup & Installation

1. Clone the repository

git clone https://github.com/yourusername/medbot.git
cd medbot

2. Install dependencies

pip install -r requirements.txt

3. Model Training & Fine-Tuning on Vertex AI

   • Dataset Preparation

     • Collected datasets:
     • PubMedQA → medical Q&A.
     • A-Z Medicine Dataset of India(Kaggle)

   • Preprocessing: Converted into instruction–response JSONL (Vertex AI requires this format):

     {"input_text": "What is the treatment for diabetes?", "output_text": "Diabetes is managed with lifestyle changes, monitoring, and medications like metformin."}
     {"input_text": "What are side effects of aspirin?", "output_text": "Stomach pain, nausea, heartburn, and bleeding risk."}
     

     Uploaded dataset to Google Cloud Storage bucket (gs://your-bucket/medbot_dataset.jsonl).

   • Vertex AI Model Selection

     • Vertex supports custom fine-tuning of models like Llama / Gemini.
     • MedBot was built on Gemini Flash 2.0 lite.

   • Model Fine-Tuning and Deployement

     • The JSONL files are uploaded in buckets for fine-tuning.
     • The model can be accessed via a CLI but for our use case we deployed it in a bucket.

4. GUI Endpoint

   1. In the gui_endpoint.py file, update the config on the code.

PROJECT_ID = "your-project-id"
REGION = "us-east1"
ENDPOINT_ID = "your-endpoint-id"
SERVICE_ACCOUNT_FILE = "/absolute/path/to/medbot-api-key.json"
IMAGE_PATH = "medbot_image.jpg"  # Optional splash image

2. MedBot had a GUI but can also be accessed on the terminal, all using the Vertex AI SDK.

3. Run MedBot

   python3 gui_endpoint.py
   

How It Works

• Voice Input → Captured via microphone (PyAudio).
• Speech-to-Text → Real-time transcription using Google Cloud Speech API.
• AI Response → Prompt sent to Vertex AI fine-tuned endpoint using the Python SDK.
• UI Display → Response shown in scrollable chat history.
• Pill Dispensing (Optional) → Raspberry Pi triggers servo motors based on detected symptoms.

Notes

• Raspberry Pi servo code is commented out for macOS; works on Linux/Raspberry Pi OS with pigpio.
• Ensure microphone permissions are enabled for real-time voice input.
• Token usage is logged in the UI (prompt, response, remaining tokens).

References

• Vertex AI Python SDK Documentation
• Google Cloud Speech-to-Text
• Tkinter Documentation
• ttkbootstrap Documentation