Feature: Integrate Neo4j for Relationship-Aware Property Recommendations

🚀 Feature Request

Add Neo4j as a graph database layer to complement MongoDB (document store) and Pinecone (vector search). This will enable EstateWise to handle relationship-heavy queries, improve explainability of recommendations, and unlock new AI/graph hybrid capabilities.

---

🔑 Motivation / Why

Current stack:

• MongoDB → user data, raw listings.
• Pinecone → semantic similarity (embeddings).
• Redis → caching.

What’s missing:

• A way to model explicit relationships (neighborhoods, schools, user preferences, property-to-property similarities).
• Ability to answer graph traversal queries that are hard/inefficient in MongoDB.
• Better explainability for recommendations: “This house was recommended because it’s in the same school district and similar in size/price to your liked properties.”

---

🎯 Use Cases

1. Neighborhood Graph

   • (Property)-[:IN]->(Neighborhood)-[:HAS]->(School)
   • Query: “Show me houses near UNC with A-rated schools and parks.”

2. Property Similarity Graph

   • (House A)-[:SIMILAR {priceDiff:20k, bedDiff:1}]->(House B)
   • Explainable recommendations, not just opaque vector matches.

3. User–Property Interaction Graph

   • (User)-[:LIKED]->(Property)
   • Collaborative filtering: “Users who liked this home also liked these.”

4. Proximity / Commute Queries

   • With Neo4j spatial: “Find 3BR homes within 15 min of downtown.”

5. Hybrid Search

   • Pinecone → fast semantic candidate retrieval.
   • Neo4j → relationship-aware filtering/ranking.

---

🛠 Implementation Plan

• Backend

  1. Install driver: npm install neo4j-driver.

  2. Add env vars:

     NEO4J_URI=neo4j+s://<db-uri>
     NEO4J_USER=<user>
     NEO4J_PASSWORD=<password>
     

  3. Create services/neo4jService.ts to init driver + expose query helpers.

  4. Define graph schema + migration script:

     • Nodes: Property, User, Neighborhood, School, etc.
     • Edges: :IN, :SIMILAR, :LIKED, :HAS.

• Data Ingestion

  • During property upsert (currently to Pinecone + Mongo), also push to Neo4j.
  • Script example: npm run upsert:neo4j.

• API Layer

  • New endpoints for relationship queries (e.g. /api/properties/graph-similar, /api/properties/by-neighborhood).

  • Extend chat agent so Decision AI can choose between:

    • Pinecone (semantic).
    • Neo4j (graph queries).
    • Hybrid (both, merged).

• Docs

  • Update ARCHITECTURE.md and TECH_DOCS.md to reflect new graph layer.

---

✅ Acceptance Criteria

• Neo4j service connected + working in dev.
• Properties + relationships ingested into graph.
• API endpoints available for graph-based queries.
• Chat agent can leverage Neo4j when user queries involve relationships.
• Updated diagrams/docs with Neo4j included in architecture.

---

📌 Notes

• Could use Neo4j AuraDB Free Tier (cloud) for dev.
• Initial focus: Neighborhood + User–Property graphs.
• Future: Pathfinding, commute queries, explainable recs.

---