Data@Urban
Explore the code, data, products, and processes that bring Urban Institute research to life.
1
A person’s financial well-being is nuanced, encompassing many different metrics and situations. A single dataset rarely paints a complete picture of people’s financial lives. Therefore, building a holistic understanding of financial well-being often requires linking data from several disparate sources.
Latest Posts
Illustration credit: Muharrem huner, Getty Images

Using Agentic AI to Create More Sophisticated Data Engineering Pipelines

- We used Claude Code to modernize NCCS's data engineering pipelines, replacing an out-of-sync, 20-repository system with single-responsibility functions, checkpointed runs, and full data provenance. Clear architectural decision records and data contracts let us keep agentic tools synchronized and auditable, helping monthly downloads grow from about 2,700 in 2024 to over 23,000 by summer 2026.
Illustration credit: Alysheia Shaw-Dansby, Urban Institute

How Linking Eviction Records to HUD Administrative Data Could Improve Housing Access

- For policymakers and housing experts to use eviction filing data to more effectively house individuals, they need to understand the key challenges, opportunities, and implications of these data. A recent project conducted by Urban Institute researchers using a grant from the US Department of Housing and Urban Development sought to meet this need by testing how stakeholders could pair eviction filings with administrative data.
An illustrated banner depicting a human head silhouette with a brain and lightbulb inside, surrounded by icons representing AI and technology concepts including gears, chat bubbles, a search bar, documents, and puzzle pieces, set against a dark blue background.

Building an Agentic AI Tool to Support Our Upward Mobility Initiative Dashboard

- We're building an agentic AI tool for our Upward Mobility Initiative Dashboard, using a custom lookup tool to reliably match place names to the right data, even in tricky cases like ambiguous city and county names. This agentic approach outperformed our earlier chatbot, handling edge cases smoothly without needing a smarter model, just better structure. We're excited to build on this as we move toward a single front door for AI-supported mobility technical assistance.
An illustrated banner depicting a human head silhouette with a brain and lightbulb inside, surrounded by icons representing AI and technology concepts including gears, chat bubbles, a search bar, documents, and puzzle pieces, set against a dark blue background.

How We Built an AI Evaluation Framework with Experts in the Loop

- We built an AI evaluation framework with domain experts in the loop to test whether our Upward Mobility Initiative knowledge base gave accurate, well-sourced answers, then used their feedback to fix retrieval and citation issues across two rounds of testing. We found the model performed well on straightforward questions but struggled with complex, multidocument ones, and that automated metrics alone couldn't replace expert judgment.
An illustrated banner depicting a human head silhouette with a brain and lightbulb inside, surrounded by icons representing AI and technology concepts including gears, chat bubbles, a search bar, documents, and puzzle pieces, set against a dark blue background.

What It Takes to Make Research and Policy Knowledge AI Ready

- We built an AI-ready knowledge base for Urban's Upward Mobility Initiative by combining careful document curation, preprocessing, and metadata tagging, and found that these steps meaningfully improved how well AI tools could ground answers in trusted research and data.