How to build a Open Data Insights AI Agent

TL;DR

This AI agent automates open data research and analysis. It interprets a user’s question, finds relevant public datasets, performs deep research, synthesizes insights, and delivers a polished, visual report—no data science skills required.

What It Does

• Understands a user’s open data question (e.g., “Homelessness in LA”)

• Finds and analyzes authoritative public datasets (Census, FBI, CDC, etc.)

• Performs deep research and trend analysis using AI

• Synthesizes findings into a clear, decision-ready report with suggested visualizations

Who It’s For

• Policy analysts, journalists, researchers, and civic tech teams

• Anyone needing fast, reliable insights from public/open data

• Non-technical users who want data-driven answers without coding

Time to Value

• Minutes: Just describe your question—get a full analysis and report in one go

• No manual data wrangling, searching, or chart-building required

Output

• A markdown report with:

  • Executive summary

  • Key trends, patterns, and comparative insights

  • Citations to data sources

  • Suggested and generated charts/visualizations

Common Pain Points for Open Data Analysis

• Finding trustworthy, up-to-date datasets

• Translating business questions into data queries

• Combining multiple sources for comparative analysis

• Creating clear, visual reports for non-technical audiences

• Time-consuming manual research and charting

What This Agent Delivers

• Automated discovery of relevant public datasets

• AI-driven research and trend analysis

• Side-by-side comparisons across time, geography, or demographics

• Clear, markdown-formatted reports with visualizations

• Citations and links to all data sources

• No coding or data science expertise required

Step-by-Step Build (StackAI Nodes)

Below is a walkthrough of each node in the workflow, what it does, and its goal:

1) User Request (in-0)

What it does:

• Collects the user’s open data question or analysis need.

Goal:

• Capture the research topic in plain language.

2) Query and Analysis Planner (llm-0)

What it does:

• Interprets the user’s request.

• Identifies relevant public datasets.

• Generates a structured plan for research and analysis (as JSON).

Goal:

• Translate the user’s question into actionable research steps and search queries.

Instructions

You are an expert open data analyst. 
  Given a user request, identify the most relevant public datasets (e.g., Census, FBI UCR, CDC, state repositories) and generate a JSON with: 
- "search_query": a concise search string for web data discovery - "research_instructions": detailed instructions for trend, correlation, and comparative analysis (across time, geography, or demographics) - "comparison_dimensions": list of dimensions to compare (e.g., city, year, demographic) 
  Return ONLY a valid JSON object, with no extra text or explanation

Prompt

User request: {in-0} 

Return ONLY a valid JSON object with the following keys: 
- "search_query" - "research_instructions" - "comparison_dimensions"

3) Web Search (action-0)

What it does:

• Searches authoritative open data portals (e.g., Census, FBI, CDC) for datasets matching the plan.

Goal:

• Find the most relevant, trustworthy data sources for the analysis.

4) Deep Research (action-1)

What it does:

• Performs in-depth research and analysis on the discovered data sources.

Goal:

• Extract deeper insights and context from the data.

5) Insight Synthesizer (llm-1)

What it does:

• Aggregates and analyzes all findings.

• Identifies trends, patterns, and comparative insights.

• Suggests and describes visualizations.

• Outputs a markdown report for decision-makers.

Goal:

• Turn raw data and research into a clear, actionable report.

Instructions

You are an open data insights assistant. Aggregate and analyze the results from the web search and deep research. Identify patterns, trends, and correlations. If comparison dimensions are present, perform side-by-side analysis. Suggest and describe visualizations (charts/maps) for the findings. Output a markdown report with:

- Executive Summary

- Key Trends & Patterns

- Comparative Insights

- Citations (with URLs)

Make the report clear for non-technical stakeholders

Prompt

User request: {in-0}



Web search results: {action-0.results}



Deep research: {action-1.content}



Analysis plan: {llm-0}

6) Analysis Tool (action-2)

What it does:

• Analyzes data, creates charts, and translates natural language analysis requests into Python code.

• Executes the generated Python code in a secure, stateless environment.

• Returns the code, results, and a public URL to any generated chart or image.

Goal:

• Automate the process of data analysis and chart creation based on the synthesized insights, making results visual and actionable.

7) Report Formatter (template-0)

What it does:

• Formats the insights and generated chart into a user-friendly markdown report.

Goal:

• Deliver a polished, readable report ready for sharing or presentation.

8) Output (out-0)

What it does:

• Presents the final report to the user.

Goal:

• Provide a single, clear output containing all findings and visualizations.