MusafirAI: RAG-Powered Travel Itinerary Generator
🧳 Project Overview
MusafirAI transforms travel planning through Retrieval-Augmented Generation (RAG) technology, creating personalized Pakistan itineraries by combining AI with curated travel knowledge. Built with LangChain and Cohere/Gemini AI, this system converts travel guides into intelligent recommendations with source attribution, delivering culturally rich travel plans.
🔧 Core Functionality
MusafirAI solves the problem of generic travel planning by implementing:
- Destination Intelligence: Semantic search across curated Pakistan travel documents
- Personalized Itineraries: Day-by-day plans based on budget, interests, and group size
- Contextual Recommendations: Culturally appropriate suggestions with practical info
- Knowledge Expansion: Upload custom travel guides to enhance recommendations
- Source Attribution: Show sources used for each itinerary component
🏗️ Technical Architecture
System Architecture Diagram
System Components
1. Document Processing Engine(document_loader.py):
- DirectoryLoader with text chunking (600 tokens, 100 overlap)
- Metadata enrichment for locations (city, province, type)
2. Vector Intelligence Core (vector_store_manager.py):
- ChromaDB/FAISS vector stores with SentenceTransformer embeddings
- Automatic persistence and recreation of knowledge base
3. RAG Orchestrator (rag_system.py):
- RetrievalQA chain with MMR search and score filtering
- Custom prompt engineering for structured itineraries
- Dual LLM support (Cohere Command/Gemini Pro)
4. Web Interface (MusafirAI.py):
- Streamlit UI with PDF export and source inspection
- Dynamic form handling for travel preferences
- Vector store management controls
Technology Stack
- AI Framework: LangChain 0.0.340
- LLMs: Cohere Command, Google Gemini Pro
- Embeddings: all-MiniLM-L6-v2 Sentence Transformers
- Vector DB: ChromaDB 0.4.15 / FAISS 1.7.4
- Web Framework: Streamlit 1.28.1
- Dependencies: Python 3.11, ReportLab, BeautifulSoup
🚀 Key Features & Innovations
Advanced RAG Implementation
- Destination-aware chunking preserving location context
- Hybrid metadata filtering (city/province/type) + semantic search
- Prompt engineering enforcing day-by-day structure with practical details
User Experience Excellence
- Interactive preference form with Pakistan-specific options
- Source document inspection with metadata
- One-click PDF/text itinerary downloads
- Real-time vector store management
Production-Ready Features
- API key management with .env and Streamlit Secrets
- Automatic directory creation for cloud environments
- Comprehensive error handling with user feedback
- Dual deployment support (local/cloud)
System Workflow
System Capabilities
- Processes 50+ page documents in under 60 seconds
- Generates 7-day itineraries with 5+ attractions/day
- Maintains context across multiple queries
- Accurately attributes sources with content previews
Validation Metrics
- Processes 50+ page documents in under 60 seconds
- Generates 7-day itineraries with 5+ attractions/day
- Maintains context across multiple queries
- Accurately attributes sources with content previews
🛠️ Implementation Details
Installation & Setup
git clone https://github.com/zshafique25/RAG_Assisstant.git
cd RAG_Assisstant
# Install dependencies
pip install -r requirements.txt
# Add API keys to .env
echo "COHERE_API_KEY=your_key" >> .env
echo "GEMINI_API_KEY=your_key" >> .env
# Launch app
streamlit run MusafirAI.py
Configuration Requirements
- Cohere or Gemini API key
- Travel documents in travel_documents/ directory
- Python 3.11 (via runtime.txt)
Prompt Engineering
PROMPT_TEMPLATE = """Create Pakistan travel itinerary using context:
Context Information:
{context}
User Query: {question}
Instructions:
1. Create day-by-day itinerary
2. Include key attractions from context
3. Add practical info: travel times, costs
4. Suggest accommodation and dining
5. Include cultural etiquette and safety tips
6. Keep concise and realistic"""
🎯 Use Cases & Applications
Tourism Industry
- Travel agencies generating personalized packages
- Hotel chains creating neighborhood guides
- Tourism boards promoting regional destinations
Traveler Applications
- Solo travelers discovering hidden gems
- Families planning multi-generational trips
- Adventure seekers finding offbeat experiences
Educational Use
- Cultural studies students exploring regions
- Geography classes analyzing travel patterns
- Language learners exploring local dialects
🔬 Technical Innovations
RAG Pipeline Optimization
- Dynamic chunk sizing based on content type
- Hybrid retrieval (similarity + metadata filtering)
- LLM fallback mechanism (Cohere → Gemini)
Knowledge Management
- Automatic metadata extraction from filenames
- User-uploaded document processing pipeline
- Vector store versioning through force_recreate
Cloud Optimization
- Ephemeral storage handling for serverless environments
- Dependency resolution for Python 3.11
- Secret management with Streamlit TOML
📈 Future Enhancements
- 🗺️ Interactive Pakistan map for destination selection
- 💬 Chat interface for itinerary modifications
- 📊 Budget calculator with real-time pricing
Long-Term Vision
- Multi-format support (PDF/Word/YouTube transcripts)
- Hotel/activity booking API integration
- Local language support (Urdu/Pashto)
- AR experience for destination previews
🏆 Project Impact & Value
Technical Contributions
- Complete open-source RAG travel assistant
- Production-ready cloud deployment configuration
- Modular architecture for easy extension
Practical Benefits
- Democratizes travel expertise through AI
- Preserves cultural knowledge in retrievable format
- Increases tourism accessibility for remote regions
Innovation Recognition
- Dual LLM support with failover mechanism
- Context-aware metadata enrichment
- Cloud-native vector store management
🎓 Learning Outcomes & Skills Demonstrated
AI Engineering
- End-to-end RAG system implementation
- Vector database management/optimization
- LLM integration and prompt engineering
Software Development
- Modular architecture with separation of concerns
- Streamlit UI development with custom CSS
- PDF generation with ReportLab
DevOps & Production
- Cloud deployment on Streamlit Community Cloud
- Dependency management for constrained environments
- Secret management and security practices
📋 Conclusion
MusafirAI represents a cutting-edge application of RAG technology to revolutionize travel planning for Pakistan. By combining semantic search with generative AI, it delivers personalized, culturally-rich itineraries grounded in verified travel knowledge. The project demonstrates professional-grade architecture with production-ready error handling, cloud deployment, and user-friendly interfaces.
The system's modular design allows easy extension to new regions or languages, while its open-source nature provides a valuable resource for developers exploring RAG implementations. MusafirAI stands as a testament to how AI can transform traditional industries while preserving cultural authenticity.
Documentation: Comprehensive setup/usage instructions in README.md
Certifiedunder the RAG-Based AI Assistant module in the Agentic AI Essentials program.