Bharat Tourism AI: An Intelligent Travel Planning Platform for India

Overview

Bharat Tourism AI is an end-to-end, production-ready travel assistant platform that combines classical machine learning, generative AI, and modern deployment practices to help users plan personalized trips across India. The goal was to move beyond a simple prototype and build something that reflects how AI systems are actually deployed and scaled in the real world.

Problem Statement

Planning a trip across India often means dealing with scattered information, unpredictable crowd levels at popular destinations, and generic recommendations that don't account for individual preferences or budget. I wanted to build a platform that could solve this using AI — combining prediction, personalization, and conversational assistance in one place.

Key Features

Crowd Prediction: A Random Forest classifier predicts crowd levels at tourist destinations, helping travelers avoid overcrowded spots.
Sentiment Analysis: A PyTorch-based feed-forward neural network evaluates traveler reviews and feedback using TF-IDF vectorization.
RAG-Powered Chatbot: A retrieval-augmented chatbot built with LangChain answers travel-related queries using semantic search over a knowledge base.
Multi-Agent Itinerary Planner: A system of coordinated AI agents generates personalized, day-by-day itineraries based on user preferences and budget.
Localized Cost Estimation: Itinerary costs are estimated and displayed in Indian Rupees (INR) for practical relevance.
User Authentication: A complete registration and login system for personalized user sessions.
Responsive UI: A light/dark mode toggle for a better user experience across devices.

Technical Architecture

Machine Learning Layer

Random Forest Classifier for crowd prediction — achieved 71.5% accuracy and a 64.2% weighted F1-score
PyTorch feed-forward neural network with TF-IDF vectorization for sentiment analysis — achieved 100% precision and recall

Generative AI Layer

LangChain framework orchestrating the chatbot logic
OpenAI embeddings for semantic understanding of user queries
FAISS (Facebook AI Similarity Search) as the vector database for fast retrieval
Multi-agent architecture for itinerary generation, where different agents handle destination selection, budgeting, and scheduling

Backend & Deployment

Flask serving as the core web application framework
Docker for containerizing the application
Kubernetes for orchestration, with replicated pods ensuring scalability and fault tolerance in a production-like environment

Tech Stack

Python · Flask · PyTorch · Scikit-learn · LangChain · FAISS · OpenAI API · Docker · Kubernetes

What I Learned

This project pushed me to work across the full AI development lifecycle — from training and evaluating classical ML models, to building generative AI pipelines with RAG and multi-agent systems, to deploying everything using industry-standard container orchestration tools. It gave me hands-on experience with the kind of end-to-end thinking required to take an AI idea from concept to a scalable, deployable product.