Compliance monitoring · Production

North Dakota University System builds a generative AI Policy Assistant on Databricks to automate compliance search

The problem

NDUS data teams spent hours manually searching through thousands of policy documents, state laws, contracts, and codes to ensure compliance, with no shared infrastructure to collaborate or scale use cases across the system's 11 institutions.

Workflow diagram · grounded in source
1
Plain English query via API
trigger
“Users prompt the LLM within the System with plain English via an API”
2
Vector Search data sync
integration
“Vector Search enables automatic data synchronization, ensuring LLM outputs are up to date”
3
LLM policy search and response
ai_action
“Over 3,000 public PDFs are now contained and synthesized in Databricks, with the LLM surfacing automatic policy search and response”
4
Results with references and links
output
“instantly generate accurate results with references, page numbers and links”
Reported outcome

NDUS reduced time to bring new insights to market from one year to six months, launched Policy Assistant which synthesizes over 3,000 public PDFs for instant plain-English policy queries, and increased team productivity.

Reported metrics
Time to bring new insights to marketfrom one year to six months
public PDFs synthesized in Policy Assistant3,000
Team productivityincreased our team productivity
Policy Assistant build timesix months
Reported stack
Databricks Data Intelligence PlatformLlama 2DBRXFoundation Model APIsUnity CatalogVector SearchMLflowSpark Declarative PipelinesMicrosoft Azure
Source
https://www.databricks.com/customers/ndus-north-dakota-university-system
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

NDUS reduced time to bring new insights to market from one year to six months, launched Policy Assistant which synthesizes over 3,000 public PDFs for instant plain-English policy queries, and increased team productivity.

What tools did this team use?

Databricks Data Intelligence Platform, Llama 2, DBRX, Foundation Model APIs, Unity Catalog, Vector Search, MLflow, Spark Declarative Pipelines, Microsoft Azure.

What results were reported?

Time to bring new insights to market: from one year to six months; public PDFs synthesized in Policy Assistant: 3,000; Team productivity: increased our team productivity; Policy Assistant build time: six months (source-reported, not independently verified).

How is this compliance monitoring AI workflow structured?

Plain English query via API → Vector Search data sync → LLM policy search and response → Results with references and links.

Related compliance monitoring cases