Overview

RePORT AI Portal helps a study team ask questions about one clinical research study without handing raw PHI to the AI assistant.

It is built for a common bottleneck: researchers need simple cohort answers, plots, and model summaries, but every request has to go through a data manager because the raw files contain identifiers. The portal keeps custody with the data team, creates a PHI-scrubbed study bundle, and lets users ask grounded questions against that published bundle.

What It Does

RePORT AI Portal:

  • loads one study from local files;

  • creates a PHI-scrubbed bundle for analysis;

  • keeps raw data outside the AI assistant’s working area;

  • opens a local chat interface for questions about the study;

  • returns counts, tables, charts, and short explanations grounded in the study files;

  • produces audit files so the study team can verify what was processed.

Who It Helps

  • Researchers get faster answers to cohort and outcome questions.

  • Data managers spend less time manually joining columns and exporting one-off spreadsheets.

  • PIs get a repeatable way to review study status, cohort counts, and analysis outputs.

  • Reviewers get a clear audit trail without needing access to raw subject data.

Typical Workflow

  1. Put the study files under data/raw/{STUDY_NAME}/.

  2. Run the pipeline or click Load Study in the web UI.

  3. The portal publishes a PHI-scrubbed bundle under output/{STUDY}/.

  4. Open the chat UI and ask questions about the published study.

  5. Use the audit files when the team needs evidence of what changed.

Privacy in Plain Language

The raw files are treated as sensitive. The assistant is meant to work from the published, scrubbed study bundle rather than the raw input files. If a user chooses a hosted LLM provider, they are responsible for confirming that their local study policy allows that provider.

For the full privacy architecture and control evidence, see PHI Architecture and IRB/Auditor Profile.

When to Use It

Use RePORT AI Portal when:

  • your team has one study to review or analyse;

  • raw files are local and access-controlled;

  • researchers need faster answers from the same study bundle;

  • the team wants audit files for review.

Do not use it as a replacement for:

  • multi-study federated analysis;

  • formal statistical review by an epidemiologist;

  • source data cleaning before the portal sees the files;

  • an imaging or DICOM de-identification pipeline.

Next Step

Start with Quick Start if the repo is already installed, or Installation if this is a new machine.