NASA PSI Agent
Scientific knowledge and data assistant for NASA Physical Science Informatics repositories.
Designers: Subhayu Sen
Developers: Muthukumaran Ramasubramanian, Nishant Panthee
An AKD-designed agent that supports semantic search, retrieval, summarization, structured extraction, and cross-study comparison across PSI datasets, documents, reports, and publications.
The PSI Agent is an AKD-designed scientific knowledge and data assistant for NASA Physical Science Informatics-style repositories. The agent supports semantic search, retrieval, summarization, structured extraction, and cross-study comparison across PSI datasets, documents, reports, and publications. Its goal is to improve data discoverability, contextualization, and usability while reducing the manual effort required for scientific literature and data synthesis. A key feature of the agent is its research augmentation workflow, combining information retrieval, knowledge synthesis, and context expansion with human-in-the-loop validation.
Inputs
- User research questions or analysis goals
- Datasets, documents, reports, and publications related to PSI
- Scientific text, tables, metadata, and numerical values
- Variables, parameters, methods, findings, and dataset descriptions
- SME-provided validation criteria and benchmark examples
Collaborative Design
The PSI Agent works collaboratively with researchers, graduate students, SMEs, data scientists, engineers, and data curators. Users define or refine the research task through guided framing, such as topic area, dataset type, variables of interest, comparison scope, domain context, and desired output format.
The agent provides structured search, extraction, summarization, and comparison support while users retain control over scientific interpretation, dataset selection, validation, and final conclusions. It is assistive rather than authoritative, helping users organize and synthesize information without replacing expert review.
Tools and data sources
WIP
Outputs
- Semantic and keyword-based search results across PSI datasets and publications
- Extracted variables, parameters, metadata, and numerical values
- Structured summaries of methods, findings, and dataset context
- Cross-study comparisons and trend summaries
- Identification of related work, cross-domain links, and research gaps
- Structured outputs such as metadata, comparisons, and insight summaries
- Benchmarking results and validation notes