🧠 Hybrid AI Travel Assistant

🚀 Overview

This project builds a travel assistant that can understand natural-language questions and respond using data retrieved from both Pinecone and Neo4j. It has been fully refactored to remove deprecated APIs and replace paid dependencies with open, modern, and cost-free equivalents.

---

Project Screenshots

Setup Instructions

1. Set up environment

   python -m venv venv
   source venv/bin/activate      # or venv\Scripts\activate on Windows
   pip install -r requirements.txt

2. Add your keys in config.py

3. Load data into Neo4j

   python load_to_neo4j.py

4. Visualize the graph (optional)

   python visualize_graph.py

   ➡ Generates neo4j_viz.html

5. Upload embeddings to Pinecone

   python pinecone_upload_sentenceTransform.py

6. Run hybrid chat

   python hybrid_chat.py

   Example query:

   Enter your travel question: create a romantic 4 day itinerary for Vietnam
   

** NOTE : Make sure neo4j is installed on local device and running in the background before running hybrid_chat_new.py or any neo4j related files.**

Key Improvements

Area	Improvement	Impact
Pinecone	Migrated to v2 SDK, updated index handling	Fixes deprecated methods, ensures compatibility
Embeddings	Switched to all-MiniLM-L6-v2	Free, fast, and accurate semantic embeddings
Model	Replaced GPT-4o-mini → LLaMA 3.3-70B Versatile	Enables open-source, high-performance reasoning
Integration	Added search_summary()	Summarizes hybrid retrieval results
Prompting	Enhanced chain-of-thought and context clarity	Improves reasoning and itinerary coherence
Visualization	Upgraded PyVis to 0.3.2	Fixed notebook=False bug for HTML export
Neo4j	Targeted vietnam-travel database	Keeps data isolated and domain-specific

---

Deliverables

Task	Deliverable
Task 1	Screenshot of successful Pinecone uploads
Task 2	Working CLI chat session using hybrid retrieval
Task 3	Documented improvements (improvements.md)

---

Acknowledgment

It served as a great learning experience in building and optimizing hybrid AI systems.