List All Projects List Graduate Projects List Senior Design Projects List REU Projects Other Undergraduate Projects
Project # 38
Improved Context for LLM Queries on Knowledge Graphs Using the Model Context Protocol
| Team Lead: | Talha Tahmid |
| Other Team Members: | |
| Faculty Advisor(s): | Dr. Ashraf Aboulnaga |
| Department: | CSE |
| Project Type: | REU |
| Title: | Improved Context for LLM Queries on Knowledge Graphs Using the Model Context Protocol |
| Abstract: | This undergraduate research thesis evaluates whether the Model Context Protocol (MCP) can broker large language model access to RDF knowledge graphs in a way that improves factual accuracy and execution reliability. Targeting multi-hop, relation-heavy questions, our system runs an MCP server over SPARQL endpoints (e.g., Wikidata and a domain knowledge graph such as Rhea), grounds entities and predicates via search tools, generates SPARQL from parameterized templates, and enforces reliability through query linting (limits and pattern constraints), dry-run validation, caching, endpoint-aware throttling, and bounded error-driven repair. We compare this graph-based retrieval pipeline against a vector-only RAG baseline on shared evaluation sets, measuring answer accuracy, consistency, latency, query success, and timeout rates. Ablations isolate MCP overhead, schema exposure, caching, and template/repair quality. Deliverables include an open-source MCP server, reusable query templates, and a reproducible evaluation harness. Preliminary results indicate grounded graph querying improves compositional accuracy with modest latency cost, providing a practical blueprint for integrating LLMs with verifiable, structured sources. |
| Poster: | View Poster #38 (PDF) |
| Judges Scoring: | Submit Evaluation (judges only) |