Natural Language Interface for Knowledge Graph Querying

LLM-based querying interface for Sangrahaka and Neo4j knowledge graphs (Sem I 2025-26)

Knowledge graphs are powerful, but they remain inaccessible to most users because querying them requires formal languages such as Cypher. This project built a natural-language interface for knowledge graph querying that lets users ask questions conversationally while the system translates them into graph operations under the hood.

The work centered on integrating an LLM-based querying workflow into Sangrahaka, a platform for annotating and querying knowledge graphs. The system combined LangChain, Neo4j, and local Ollama-hosted models to support privacy-preserving graph interaction, structured responses, and interactive graph visualization.

The project also explored RAG-style enhancements over graph-structured data, with the longer-term goal of making curated knowledge graphs easier to access for non-technical users in research and educational settings.

Team: Saaketh Dataram (Sem I 2025-26)