DiabetesDietBot is an AI-powered chatbot developed by Syed Faizan for Mysore Medical College and Research Institute as part of a research study on how chatbots can assist doctors and medical students in recommending predefined diet plans to patients with Type 2 Diabetes.

The bot leverages Retrieval-Augmented Generation (RAG), FAISS vector search, and OpenAI’s GPT-3.5-turbo to provide personalized diet recommendations based on patient profiles, such as:

• Age (e.g., 40s, 50s)
• Dietary Preferences (e.g., Vegetarian, Non-Vegetarian)
• Caloric Requirements (e.g., Low, Moderate, High)

It uses a dataset of structured South Indian meal plans extracted from a PDF database to generate optimized diet recommendations.

• ✅ Personalized Diet Plans – Generates meal plans tailored to user inputs.
• ✅ RAG-Based Search – Retrieves relevant meal plans using FAISS for similarity search.
• ✅ Gradio UI – Interactive chatbot interface for ease of use.
• ✅ Secure OpenAI API Integration – Uses GPT-3.5-turbo for intelligent responses.
• ✅ Customizable Meal Database – Works with preloaded PDF meal plans.
• ✅ Optimized for South Indian Diets – Designed for diabetes-friendly traditional meals.

📦 DiabetesDietBot
├── 📜 app.py # Main Gradio-based chatbot UI
├── 📜 rag.py # Retrieval-Augmented Generation (RAG) implementation
├── 📜 rag2.py # Alternative FAISS-based text processing
├── 📂 db_mealplans # FAISS vector store for meal plan retrieval
├── 📜 .env # Environment variables (OpenAI API key)
└── 📜 README.md # Project documentation (this file)

git clone https://github.com/your-repo/DiabetesDietBot.git
cd DiabetesDietBot

Ensure you have Python 3.8+ installed. Then run:

pip install -r requirements.txt

Create a .env file in the project root and add your OpenAI API key:

OPENAI_API_KEY=your_openai_api_key

The meal plans are extracted from MealPlans.pdf using PyPDF2:

from PyPDF2 import PdfReader
reader = PdfReader("MealPlans.pdf")
text = "\n".join([page.extract_text() for page in reader.pages])

The extracted text is embedded using OpenAIEmbeddings and stored in a FAISS vector database:

from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import FAISS

embeddings = OpenAIEmbeddings()
vector_store = FAISS.from_texts([meal_plans_text], embedding=embeddings)

When a user inputs a query, FAISS retrieves top 3 most relevant meal plans:

docs = vector_store.similarity_search(user_input, k=3)
meal_suggestions = '\n\n'.join([doc.page_content for doc in docs])

A user-friendly Gradio-based chatbot interface allows interaction

The chatbot has a minimalist UI with meal plan visualization.

Users can input preferrences And receive a tailored meal plan optimized for Type 2 Diabetes.

The Diabetes Diet Bot was rigorously evaluated using RAGAS (Retrieval-Augmented Generation Assessment Score) metrics to assess the relevancy, correctness, and semantic similarity of the responses generated by the model.

• High Relevancy: The bot consistently provides meal plans tailored to the user’s requirements, maintaining high relevancy across all queries.
• Strong Semantic Similarity: Responses closely align with ground truth, reflecting accurate understanding of dietary needs.
• Areas for Improvement: Slight dips in correctness for high-caloric vegetarian plans suggest potential enhancements in nutrient-specific recommendations.

• 🔹 Expand Meal Database – Include more regional & international diabetes diets.
• 🔹 Nutritional Analysis – Provide macronutrient breakdown for recommended meals.
• 🔹 Voice Interface – Integrate with speech recognition for accessibility.
• 🔹 Doctor’s Portal – Allow medical professionals to curate and modify diet plans.

Query: “What is the meal plan for a Male in his 20s who is vegetarian with high caloric needs?”

Response:

Day 1:

• Breakfast: Ragi porridge with mixed nuts and dates, banana
• Lunch: Brown rice, sambar, mixed vegetable curry, curd, roasted peanuts
• Snack: Chana chaat, coconut water
• Dinner: Whole wheat dosa, palak dal, cucumber raita

The Diabetes Diet Bot demonstrates exceptional capability in generating personalized meal plans with high relevance, factual correctness, and semantic precision. Its performance makes it a reliable assistant for dietary consultations, particularly for diabetic and health-conscious individuals.

Note: Evaluation results were generated using RAGAS and logged with Weights & Biases (WandB) for comprehensive performance tracking.

This project is open-source and licensed under the MIT License.

🛠 Dr. Syed Faizan
📍 Mysore Medical College & Research Institute
🔗 LinkedIn

For collaborations or research inquiries, feel free to reach out!