Proposal: Add Experimental GraphRAG Skill

Summary

I'd like to propose a new experimental skill focused on GraphRAG architectures and retrieval patterns on Databricks.

Motivation

The repository currently contains skills covering:

• Apps
• Jobs
• Pipelines
• Lakebase
• Model Serving
• Unity Catalog (open PR)
• Vector Search (promotion PR)

There is already guidance around embeddings, vector similarity search, and RAG-related building blocks. However, I could not find any guidance covering Knowledge Graph or GraphRAG patterns.

As enterprise AI workloads mature, many teams are evaluating:

• Traditional RAG
• Hybrid Retrieval
• GraphRAG
• Knowledge Graph augmented search

and need guidance on when to use each approach.

Proposed Skill

experimental/databricks-graphrag

The skill would help users:

1. Determine whether GraphRAG is appropriate.

2. Compare GraphRAG vs traditional RAG.

3. Design retrieval architectures.

4. Combine vector retrieval with graph relationships.

5. Integrate with existing Databricks capabilities such as:

   • Unity Catalog
   • Vector Search
   • Model Serving
   • Databricks Apps

Example Topics

• Knowledge graph construction
• Entity extraction
• Relationship modeling
• Hybrid retrieval patterns
• Graph-enhanced context generation
• Evaluation considerations
• Common implementation pitfalls

Question

Would a GraphRAG-focused experimental skill fit the goals and scope of this repository before I begin implementation?