AI Lab
AI that does the data work, with your staff in charge
Three jobs ARC staff do by hand today, run by AI on the region's real data. AI makes the judgement calls, built-in rules check every answer and do all the arithmetic, and a person approves the result before anything is published.
Time estimates are our estimates for typical regional-agency workflows.
The problem
Every quarter, permit exports arrive from dozens of cities and counties, each with its own columns and spellings: "Cobb Co. Comm. Dev.", "D'ville", "PTC". Coordinates are swapped, unit counts are blank, permits are duplicated. Staff clean them by hand before the Housing dashboard can count production against the 367,000-home goal.
Try it
Use our sample: a quarter of synthetic permits from 30 jurisdictions with typical problems built in. Or upload your own CSV or Excel export.
Data steward's summary
Approve and publish
Publishing adds three indicators (all units, single-family, multifamily) to the explorer for every tract and county, through the normal admin rebuild. It needs an admin sign-in.
Which column is which
Jurisdiction names matched
Permits placed in census tracts
Housing units by county
Every fix and flag
The problem
Planners need figures for areas that don't follow census tract lines: transit station areas, Tax Allocation Districts, LCI areas, a new study area. Each one means days of GIS overlay work, and choosing the wrong method (adding up a rate, averaging a median) quietly gives wrong numbers.
Try it
Indicators (6)
Estimates by area
Method for each indicator AI proposes · rules check
Accuracy check against the published county figures
The same method run on county boundaries, compared with the county figures ARC already publishes.
Results
Method note for the metadata page
The problem
Board briefings, public updates and county briefings repeat the same figures in different words, every quarter. Copying numbers by hand from dashboards into documents is slow and is where mistakes creep in.
Try it
Draft edit it here
Chart
Figures given to the AI
Every number in the draft must come from these, and is checked.
The problem
Residents, elected officials and partner agencies email ARC staff questions the data could answer, if they knew where to look.
What it does
Ask the data is on the explorer and both dashboards. It looks every figure up in the published data with read-only tools, checks each number in its answer, cites the Census tables it used, declines questions outside the data, and offers to show the answer on the map.
Try it
Ask the data needs AI to be on for this server.
Use the same data from Claude, Copilot or any MCP-compatible assistant
The prototype includes an MCP server (tools/mcp_server.py) that gives AI assistants the same read-only lookups:
search indicators, county tables, rankings, threshold counts and area profiles. Staff can ask questions of ARC's data from the assistant they already use. To connect Claude Desktop:
{
"mcpServers": {
"arc-regional-data": {
"command": "python3",
"args": ["/path/to/arc-regional-data/tools/mcp_server.py", "--url", "https://arc-demo.mindedgesolution.com"]
}
}
}