From Raw Data to Interactive Results: AI-Assisted Data Workflows

Session Submission Type: AI Peer Learning Lab

Abstract

Automating Text Analysis using Large Language Models
Facilitator: Zachary Courser - Claremont McKenna College, Professor of the Practice; Director, Policy Lab
Training Category: Qualitative Research
AI Tool: ChatGPT
Participant Type: Researchers
Skill level recommended: Intermediate
Recommended preparation or pre-work before session: Familiarity with qualitative social science methodologies, particularly content analysis and other uses of text as data, is a prerequisite. Experience with Python, R, and other programming languages is not necessary, but will aid understanding of how to impliment a retrieval augmented generation (RAG) pipeline. Lastly, familiary with the use of an application programming interface (API) is also recommended.

This tutorial will focus on retrieval-augmented generation (RAG) pipelines as a general approach to automated text analysis in the social sciences that harnesses their national language processing (NLP) abilities. It is intended to introduce social scientists to emerging methods for using artificial intelligence in text analysis research. The design, implementation, and evaluation of AI-automated workflows for research tasks such as text classification, qualitative coding, and data extraction will be explored. The promise of large language models (LLMs) to advancing social science research lies in researchers’ understanding of their probabilistic nature, the ability to control parameters, evaluate performance, and demonstrate the validity of results. These topics will be integrated into the discussion of creating a RAG pipeline and evaluating the validity of its outputs.


******
Building a novel dataset using AI assisted web scraping
Facilitator: Shalini Mustala - Public Policy Institute of California, Research Associate
Training Category: Research Productivity & Workflow Automation
AI Tool: Claude code
Participant Type: Researchers
Skill level recommended: Intermediate
Recommended preparation or pre-work before session: Participants are recommended to install Python and Claude code on their computers

This session will explore how researchers can incorporate AI into complex data-collection and analysis workflows while maintaining transparency, reproducibility, and opportunities for human review. Drawing on a project to identify school wellness centers across California, the session will walk participants through the development of a pipeline that identifies and downloads Local Control and Accountability Plans (LCAPs) across school districts and extracts information about wellness centers from hundreds of documents. The project illustrates a hybrid approach in which deterministic methods, web scraping and AI are used selectively rather than relying on AI for every task. The underlying workflow prioritizes cheaper and more reproducible methods first, uses tiered fallbacks when websites differ, and records failures. The downstream document analysis similarly favors traditional text processing methods over AI-generated synthesis when reproducibility is more valuable. The training will use a “Build the Workflow” activity. Participants will work with a simplified research scenario involving information scattered across websites and PDFs. They will map a workflow and decide which tasks are best handled through traditional code, AI/LLMs or human review. We will then discuss tradeoffs involving accuracy, scalability, cost, and reproducibility. By the end of the session, participants will be able to identify appropriate roles for AI within research workflows, design fallback and validation strategies, and recognize situations where conventional methods or human review may be preferable to AI.

******

Turning Excel Charts into Interactive Data Visualizations using AI
Facilitator: Jonathan Schwabish - Urban Institute, Senior Fellow
Training Category: Data Visualization, Graphics, Presentations & Web Content
AI Tool:  Claude
Participant Type: Researchers
Skill level recommended: Beginner
Recommended preparation or pre-work before session: Participants should bring a laptop with Excel and access to a generative AI tool. Participants should also take a few minutes to explore the Tableau Public website to get a broad sense of what Tableau dashboards often look like: https://public.tableau.com/app/discover. 

Excel is still the tool that most analysts, researchers, and communicators build their charts. The results are usually static images pasted into reports and slide decks. Meanwhile, readers increasingly expect to hover, filter, and explore data on their own, and other researchers often want to get their hands on the data. Until recently, closing that gap meant learning JavaScript libraries such as D3 or Chart.js but generative AI tools have changed that. In this hands-on session, participants will learn a practical workflow for turning an Excel chart into an interactive, web-ready data visualization using AI assistants, without writing code from scratch. The session opens with a brief live demonstration. Starting from a standard Excel dataset, I’ll use Claude AI to generate an interactive HTML version with tooltips, filter, and annotations. Along the way, I'll show what the AI does well, where it stumbles, and why the human in the loop still matters. I’ll also demonstrate some additional projects and tools I’ve built using Claude, noting the things it does well and the things it sometimes misses. The hands-on portion forms the core of the session. Working individually or in small groups, participants will receive a sample dataset and use the AI tool of their choice to build an interactive version. A short “spot the error” exercise asks participants to find mistakes in the AI-generated chart, reinforcing the importance of verification and factchecking. We’ll close by sharing a few participant results, comparing the prompts that produced them, and discussing when interactivity genuinely helps readers and when a well-designed static chart is the better choice. By the end of the session, participants will be able to prepare Excel data and charts for AI-assisted conversion; write clear, specific prompts that produce usable interactive visualizations; evaluate AI-generated output for accuracy, clarity, and accessibility; and decide when interactivity adds value for their audience. Participants should bring a laptop with Excel and access to a generative AI tool. No programming experience is required.

Policy Area

• AI Peer Learning Lab

Facilitators

• Zachary Courser, Claremont McKenna College

• Shalini Mustala, Public Policy Institute of California

• Jonathan Schwabish, Urban Institute

Individual Presentations

• Automating Text Analysis using Large Language Models - Facilitator: Zachary Courser, Claremont McKenna College

• Building a novel dataset using AI assisted web scraping - Facilitator: Shalini Mustala, Public Policy Institute of California

• Turning Excel Charts into Interactive Data Visualizations using AI - Facilitator: Jonathan Schwabish, Urban Institute