NASA builds people knowledge graph using graph tech and AI

NASA is leveraging graph technology and large language models to transform how it connects its most valuable resource—its people. By building a People Knowledge Graph, the agency can now dynamically map relationships between employees, skills, and projects across its vast organization. This innovative approach allows NASA to discover hidden expertise, identify skill gaps, and enable more effective collaboration across traditionally siloed centers, ultimately accelerating mission-critical work through better utilization of its human capital.

The big picture: NASA’s People Analytics team has created a knowledge graph that connects workforce data, skills, and projects to enable better talent discovery and organizational insights.

• The graph database approach overcomes limitations of traditional relational databases by directly modeling connections between people, their skills, and their work.
• By combining graph technology with LLMs and secure AWS infrastructure, NASA can now answer complex questions about expertise and organizational capabilities that were previously difficult to address.

Key components: The solution runs on NASA’s secure internal AWS cloud using Memgraph database technology alongside an on-premises LLM deployment.

• Memgraph runs in Docker on EC2 instances while an on-premises LLM server (Olama) handles skill extraction and chatbot querying.
• AWS S3 buckets store both structured and unstructured data, with GQLAlchemy facilitating data ingestion from S3 to Memgraph using Cypher queries.
• The system’s architecture balances security requirements with the need for powerful computing resources to handle complex graph operations and language model processing.

Data integration approach: The team combined multiple internal data sources to build a comprehensive view of NASA’s human capital.

• People data came from NASA’s internal Personnel Data Warehouse, while AI/ML project information was sourced from the AI Use Case Registry.
• Skills were extracted from team resumes using language models, creating a more complete profile of employee capabilities beyond formal job titles.

Graph modeling strategy: NASA built a labeled property graph with nodes representing employees, their attributes, skills, and organizational relationships.

• The graph includes nodes for employees, position titles, occupation series, pay grades, organizations, projects, education levels, universities, instructional programs, majors, and extracted skills.
• This comprehensive schema allows for multidimensional analysis of the workforce, revealing connections that wouldn’t be visible in traditional databases.

Current limitations: While powerful, the knowledge graph is still in development with approximately 27,000 nodes and 230,000 edges.

• Future plans include enhancing data quality, automating the data pipeline, and adding employee learning goals to the graph.
• The team aims to improve Cypher query generation and retrieval-augmented generation accuracy while scaling up to over 500,000 nodes and millions of edges.

Why this matters: The People Knowledge Graph represents a shift in how large organizations can understand and leverage their workforce capabilities.

• By mapping skills, experience, and relationships in a graph structure, NASA can more effectively match talent to mission needs and identify expertise gaps.
• This approach could serve as a model for other government agencies and large organizations seeking to better understand and utilize their human capital resources.