Strategic Initiative Fund (SIF)

Mi DuRAG: Human-Centered AI for Data Analysis in Research and Education

A central challenge in AI development is getting AI to work with people instead of replacing them. Most AI data tools treat analysis as something to fully automate: upload a file, get an answer, then start over next time. Michigan Dual-layer Reasoning and Analysis Graphs (Mi DuRAG) aims to create a new approach for AI-driven data analysis that works with people rather than in place of them — an AI data partner that captures expert decisions, rationales, and data structure in a persistent graph rather than fleeting chat logs.

At its core, the project develops a principled approach to capturing expert decisions and rationales throughout the analysis process, so that reasoning can be reused by collaborators, students, and future researchers, advancing both discovery and the way research is taught. Mi DuRAG links messy research spreadsheets to the knowledge of how they should be used, so the system can handle documentation, transformations, and organization while students and researchers focus on analytical questions. The project combines anthropological research on human biological variation with computer science research on self-optimizing database systems.

Mi DuRAG will produce:

• a public benchmark for AI-assisted research data analysis on messy biological, laboratory, and survey datasets
• a dual-layer reasoning and analysis graph architecture and open-source prototype tool
• controlled evidence on student learning and research handoffs
• peer-reviewed publications on the architecture, benchmark, and educational study in leading databases, AI, and computing-education venues