Public Data Desk

Ask a question about public data.
Get an answer — and the code — back.

The data's public and the tools exist — you just need someone to point them at your question. You don't have to be an expert to get a real answer, and an expert can check it fast, because the code comes back with it.

Bring us
  • Questions about open, public datasets
  • Workforce, contracting, jobs, spending, budgets
  • Things you're fine seeing analyzed in the open
Do not send us
  • Your office's private or internal data
  • Casework or constituent data
  • Anything sensitive — keep it off our desk

What becomes open source is the code, not you. We can generalize it so it's not your exact question, and your name and office never get attached.

The kinds of things you can ask

What's happening to the VA workforce — and to nurses specifically?
Which job series is the government hiring most this quarter?
Are hiring freezes showing up in the new-hire numbers yet?
Who are an agency's biggest contractors, and is spend up or down?
Does this agency buy on best value or lowest price?
For a budget account: how much was enacted vs actually spent?

What an answer looks like

Real numbers, pulled from public data for this demo. (I built the EHRI HuggingFace repo these come from, and I work with the others professionally.)

VA workforce · EHRI data

Over the last six months VA headcount kept drifting down to 446,735 — about 35,147 below its early-2024 peak. Nurses are roughly flat near 119,500, with hiring just ahead of departures lately.

VA contractors · USASpending / MakeGov

VA contracting in Minneapolis (Minnesota's 5th) rose $117M$135M. But zoom out to the state and national networks like Optum ($21.2B) dwarf local vendors — you only know that if you read the query.

How it works

1
You ask
A plain-English question about public data.
2
We pair you
A volunteer (later: paid) + an AI agent that writes the analysis.
3
You get an answer
Quick, sourced, with a chart where it helps.
4
And the code
Line by line, open source — so you can re-run and check it.

One catch, and it's the important one: understand what a number actually measured and what it meansbefore you cite it — not the code itself, but what it did. The whole method's written down, so someone who knows the topic can check it in a few minutes.

Every answer is a saved, runnable artifact

Fully in code
No spreadsheet magic — the whole analysis is a script.
Transparent
Every number traces to a public source you can check.
Reproducible
It's a script — anyone can re-run it and get the same result.
Saved
We keep it, so the next person builds on it.

And it has to be fast — you need the number before the hearing, not next week.

Get involved

This is early — I'm testing whether the model works. The near-term goal is small: about five real questions from staffers, enough to see if this helps. If you want to run with itand own it, great. Otherwise I'm looking for people to answer questions, build the site, or help fund it once it's proven — or just bring a public-data question and be an early user.

abigail.haddad@gmail.com