HELPR LLM Tools

Our HELPR (Health data Exploration Powered by AI for Research) Tools leverage AI to improve the way researchers interact with health data.

Finding, cleaning, and using health data for research is complicated. Our suite of HELPR LLM Tools make it eaiser for our AI health researchers to interact with data.

Jump to a HELPR Tool:
Cohort Builder Assistant | Query Builder Assistants

Cohort Builder Assistant

The Cohort Builder Assistant helps researchers identify patients who meet inclusion criteria for research studies.

Identifying patients fitting the criteria for clinical trial recruitment is a time-intensive and operationally burdensome component of clinical studies. Manual schedule screening, fragmented structured data filters, and limited access to outpatient notes slow recruitment and delay trials.

The Study Cohort Builder LLM solves this by:

  • Combining structured filters with semantic note analysis

  • Leveraging LLMs to extract eligibility criteria from unstructured clinical notes

  • Providing a user-friendly interface define inclusion and exclusion criteria through natural language interactions

  • Reducing expensive coordinator time spent manually reviewing charts.

For more information, contact:
Cinzia Smothers, Director,
AI & Digital Health Research Services

Query Builder Assistant

This LLM will build a SQL query that you can run in the Deidentified RDW or the RDW to find the data you’re looking for.

The DeID RDW Query Builder Assistants are powered by UMGPT and designed specifically for querying Michigan Medicine patient data. The Query Builder Assistants translate natural-language research questions into DeID RDW or RDW-ready SQL queries in seconds, expanding data access to our researchers and accelerating their workflows.

Features include:

  • Generates SQL code based on natural language prompts.

  • Automatically identifies the correct RDW tables, joins, filters, and code sets required.

  • Understands the DeID RDW and RDW schemas.

  • Trained on our data documentation, there is no need to explain the table structures, relationships, or data types. Users can get started with their research question immediately.

  • Supports iterative prompting.

  • Remembers chat history, so users can ask to adjust inclusion/exclusion criteria or output formats on the fly.

For more information, contact:
Cinzia Smothers, Director,
AI & Digital Health Research Services

Return to Top of Page