🧠🌍 AI Travel Planner: Neurosymbolic Conversational Assistant

A powerful Conversational AI that blends neural LLMs, symbolic reasoning, and a knowledge graph to generate personalized, intelligent travel itineraries from natural language queries.

Built using:

• 🤖 Ollama + LangChain for LLM-powered text generation
• 🧠 Symbolic rules for logic-based reasoning
• 🧬 Neo4j for knowledge graph trip storage and queries
• 🧵 Streamlit for a beautiful interactive UI
• 🌍 SpaCy for NLP and entity extraction
• 📊 Hugging Face Datasets for real-world trip data

---

📦 Dataset

This app is powered by the osunlp/TravelPlanner dataset available on Hugging Face.

It contains:

• Realistic trip planning queries
• Origin/destination cities
• Duration, budget, constraints
• Annotated plans: attractions, transportation, food, stay info
• Reference information about cities, flights, hotels, restaurants

Used as the core knowledge base, loaded into Neo4j as a symbolic graph.

---

💡 Key Features

• 🧾 Understands natural queries like:
  "Plan a 3-day trip from Kansas City to Pensacola with $1500 budget"

• 🧠 Extracts origin, destination, budget, duration using SpaCy + regex

• 📖 Applies symbolic reasoning (rules based on city, budget, trip length)

• 🔄 Fetches structured trip data from Neo4j (used as retrieval context)

• 🤖 Uses Ollama (LLaMA 3.1 or other local models) to generate full itineraries

• 🧩 Combines both neural (LLM) and symbolic (rules + knowledge graph) reasoning

• 🧵 Built entirely in Streamlit

---

🔧 How It Works

1. User inputs a free-text travel query
2. SpaCy extracts origin, destination, days, budget
3. Symbolic module infers trip type and travel advice using rules
4. Neo4j is queried for relevant real-world trip plans
5. LLM (Ollama) receives all structured info and generates a custom itinerary
6. Itinerary includes: transportation, accommodation, meals, attractions, and estimated costs

---

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/your-username/neurosymbolic-travel-planner.git
cd neurosymbolic-travel-planner

2. Install Dependencies

pip install -r requirements.txt
python -m spacy download en_core_web_lg

3. Start Ollama

Make sure you have Ollama installed and a model like llama3 or qwen2.5 downloaded.

ollama run llama3.1

4. Load Neo4j and Import TravelPlanner Data

Ensure Neo4j is running on your machine or remote.

Create a .env file or input connection details from the sidebar.

Use provided scripts (or your own) to load the osunlp/TravelPlanner dataset into Neo4j as a trip graph.

5. Run the Streamlit App

streamlit run travel_planner.py

🧠 Symbolic Component

We apply structured, rule-based logic for:

• Inferring trip types (e.g., city, beach, mountain) based on destination
• Budget tier classification
• Suggesting travel advice based on trip length
• Fallback logic when Neo4j returns no trips
• Built with flexible Python rules, and optionally enriched via Neo4j or LLM fallback.

🤖 Neural Component (LLM via Ollama)

• Uses Ollama to run models like llama3.1, qwen, or mistral
• Prompt includes:
  • User query
  • Extracted symbols
  • Symbolic insights
  • Retrieved trips from Neo4j
• Generates a day-by-day itinerary with natural language and cost summaries

Demo

neuro-symbolic-demo.mp4