Evidence to Accreditation: Using LLMs for Automated Evidence-Grounded Drafting

Academic Mentor | Anuj Tiwari
Community Partner | Champaign-Urbana Public Health District (CUPHD)

Project Description
Public health accreditation and reaccreditation require health departments to complete detailed documentation forms and link each response to supporting evidence such as policies, standard operating procedures, quality improvement artifacts, meeting minutes, and program reports. This process is often time-consuming, as staff must repeatedly locate documents, extract relevant details, and draft narratives across multiple forms aligned with accreditation standards.

This project explores how large language models (LLMs) can be used responsibly to support accreditation documentation through a Retrieval-Augmented Generation (RAG) system. The proposed approach ingests agency-provided evidence documents, retrieves relevant excerpts for each accreditation form section, and generates draft responses that are explicitly grounded in the source materials. The system is designed with guardrails that require evidence for every claim and route all outputs to human reviewers for verification and final edits.

Working with the Champaign-Urbana Public Health District, the project aims to develop and evaluate a prototype that reduces administrative burden while improving traceability, consistency, and audit readiness in accreditation workflows. Findings will inform how AI tools can be applied thoughtfully in public health administration to support accuracy, accountability, and efficiency.

Role of the Community-Academic Scholar:
The Community-Academic Scholar will serve as the primary developer of the project prototype, working closely with the academic mentor and community partner. The scholar will design and implement an end-to-end pipeline that converts evidence documents into draft-filled accreditation forms, including developing data schemas for forms, building document ingestion workflows, and implementing retrieval-augmented generation to produce evidence-grounded draft responses.

The scholar will also develop features that link each drafted response to its supporting evidence and flag missing or insufficient documentation through “no-evidence, no-claim” guardrails. In addition, the scholar will evaluate the system by assessing accuracy and faithfulness to source documents, completeness of drafted fields, time savings, and reviewer acceptance.

Through this role, the scholar will gain hands-on experience in applied machine learning, responsible AI system design, and community-engaged research focused on real-world public health workflows.