AI Census Data Generation Agent

How Cassidy automates census data generation using AI

Step 1: Define schema and control requirements

The Workflow begins when you specify target geographies, population universes, and the household- and person-level attributes you need—Cassidy stores these parameters in your Knowledge Base for consistent reuse.

Step 2: Acquire and normalize source data

Cassidy connects to ACS PUMS microdata, tract-level control tables from NHGIS or the Census API, and TIGER/Line geographic files, then harmonizes codes, categories, and geocodes into a unified dataset.

Step 3: Construct and validate control arrays

The Workflow builds single- and multi-way marginal control tables for each tract, checks internal consistency, and handles structural zeros and differential-privacy noise from 2020+ releases.

Step 4: Run synthesis algorithms

Cassidy executes iterative proportional fitting (IPF) or hierarchical IPU to weight seed records against tract controls, then integerizes fractional weights using controlled rounding or linear programming to produce discrete households and persons.

Step 5: Generate quality reports and deliverables

The Workflow calculates fit diagnostics (TAE, SRMSE, RSSZ), flags tracts that exceed tolerance thresholds, and packages structured microdata files (households.csv, persons.csv) with full metadata and provenance documentation.

Step 6: Human-in-the-loop review

Before final delivery, analysts review flagged tracts and validation summaries, approve or adjust parameters, and Cassidy logs every decision for reproducibility and audit.