Speakers
Description
To filter data, users need to know dataset structures, variable names, and valid value ranges. The cohortBuilder R package offers a common API for multi-step filtering across data frames, databases, and custom backends. The shinyCohortBuilder package adds an interactive Shiny GUI on top of it.
We introduce a metadata layer in cohortBuilder that connects filtering pipelines to large language models (LLMs) through tool calling. Users describe their data once with annotations and statistical summaries. Both the application and the LLM can then discover filters, read their constraints, and apply them without manual setup.
We expose cohort operations as LLM tools. These tools retrieve filter metadata, set up filtering steps, and apply values within valid ranges. The LLM turns natural language requests into filtering actions, so users do not need to (but still can) work with data schemas directly.
In shinyCohortBuilder, a chat panel lets users talk to the LLM. The LLM actions update the GUI in real time: filters appear, values change, and results refresh as the model responds.
This approach follows FAIR principles. The metadata layer makes filter definitions Findable and Accessible to both humans and machines. The source-agnostic design ensures Interoperability across different backends. By describing data once, LLMs can provide context-aware insights without users needing to learn the data model.
During the presentation, we will show a live migration from a standard cohortBuilder setup to one with LLM-driven exploration, where users interact with their data through natural language.
If you used AI tools or services to support the preparation of this submission, please state the name and reason for using each of them.
Claude 4.6 Opus for proofreading only.
| Keywords: Please list up to 5 keywords to help us find the right session for your contribution. | cohort building, LLM, FAIR data, metadata-driven filtering, Shiny |
|---|---|
| Virtual Option | This submission is for onsite presentation only |
| Video Recording | Video sharing is fine |
| The author(s) agree(s) to take responsibility and be accountable for the contents of the submission and is/are authorized to present it. | Confirm |