c0mpiled_
← back to articles
hack·april 27, 2026·15 min read

c0mpiled-10: AI for Government in DC

by truc phan351 reads

Overview

  • The DC data community had been asking for a serious AI hackathon in the District for a while. On Friday April 24, we ran one — themed AI for Government — alongside Deep Learning Adventures, Data Community DC, and Prefect. Per George Zoto, the largest AI hackathon DC has seen to date.
  • 150 registered, 125 approved, 70 checked in. 32 submissions, 3 winning teams.
  • The theme riffed off three YC Requests for Startups, plus an open prompt for participant-defined ideas. (Full participant brief here.) Build for one of the following tracks in four hours.
  1. Intake & Applications (Tom Blomfield) — AI tools to help agencies process forms and applications still handled on paper.
  2. Fraud & Investigations (Garry Tan) — Software to accelerate qui tam / False Claims Act work — parsing evidence, tracing entities, packaging complaint-ready files.
  3. LLMs vs. Gov Consulting (Gustaf Alstromer) — LLM software to replace what firms like Deloitte and Accenture currently do for government ($100B+/year market).
  4. Or surprise us* with your own angle.
  • At 9:05pm, the first submission came in — Ethan Nguyen with ATO Copilot, TurboTax for government security approvals. Solo team. Voice-AI-agent walkthrough as the demo. 10/10 from judges.
  • Big shoutout to George — "the proof-of-concept gap is real, but tonight is about closing it."

Reflections

Good:

  • Four hours vs. our usual three. People had enough time to build.
  • People arrived early. Doors opened at 4:30 but the room was already filling at 4; ~50 people in by 5pm. We aimed for 60 max, yet 70 showed up!
  • No formal mentoring sessions. However, judges were encouraged to walk the floor during build time and give feedback to teams. With majority of attendees being students or recent grads, that hands-on guidance from judges shaped a lot of the projects.
  • Strong judging panel with government expertise — five YC founders/operators (Closure, Vulcan, Crustdata, Capitol AI) plus Jordan Eccles, Presidential Innovation Fellow at the White House Council on Environmental Quality.
  • Wide range of team backgrounds. For example, Holocron Investigative came in with years of experience tracking illicit fentanyl trafficking, while Clawback walked in cold to the qui tam / FCA whistleblower process and built from scratch in the room. Both shipped credible submissions.
  • Positive feedback from participants on how organized the event was.
  • While judges deliberated in separate rooms, the rest of the teams kept presenting their projects to each other to share their projects and learn from the others. For finale, top 5 teams had an opportunity to pitch live in front of the judging panel in order to choose top 3.
  • Two food rounds — banh mi at 4:30, pizza + salad at 7:30 — plus snacks and drinks throughout. People really liked our food.
  • Convenient location, central DC, five minutes from the Metro.
  • Partnered with Data Community DC, which helped on marketing and logistics.

Bad:

  • Venue wifi was slow. Several teams fell back to their phones to tether; we ran two organizer hotspots to share. Submissions came in 15 minutes late, the event ended later than planned, and we missed a clean group photo at the end.

Observation:

  • Track distribution — Fraud & Investigations pulled 47% of submissions, Intake & Applications 31%, LLMs vs. Gov Consulting 16%. The 1st and 2nd place winners came from LLMs vs. Gov Consulting (the smallest track); 3rd was a participant-defined Local & state legislation track.
  • Team sizes — 15 of 32 submissions were solo builds, 8 were duos, 3 were trios, 4 were teams of four, 2 were teams of five. With AI in the loop, one person can build great stuff in four hours: the 1st place winner was a solo, submitted first at 9:05pm, pitched with a voice agent, and scored 10/10.

Sponsors & Partners

Judges

Prizes & Recognition

1st Place ($3,000): ATO Copilot — TurboTax for government security approvals. Solo build. First submission. Pitched with a Google Vids–generated walkthrough. 10/10 from judges.

2nd Place ($1,500): TerraTrace.ai — Replaces the 2.5-year, $2M Environmental Impact Statement process with an AI-native pipeline.

3rd Place ($500): BillTrax — Legislative software for centralizing bill tracking across local, state, and federal levels.

Submissions by track

  • Fraud & Investigations (Garry Tan RFS) — 15 (46.9%)
  • Intake & Applications (Tom Blomfield RFS) — 10 (31.3%)
  • LLMs vs. Gov Consulting (Gustaf Alstromer RFS) — 5 (15.6%)
  • Other (participant-defined) — 4 (12.5%)

All Submissions

TurboTax for Government Compliance (LLMs vs. Gov Consulting) — 1st place

TerraTrace.ai (LLMs vs. Gov Consulting) — 2nd place

BillTrax (Local and state government + industry + and lobbyists those with interest to track bills) — 3rd place

CaseBrief AI (Intake & Applications)

Casework AI (Intake & Applications + Fraud & Investigations)

CivicLedgers (Auditing Tool for Local Government Accountants)

CLAIMGUARD AI (Fraud & Investigations)

ClaimTracer (Lucas & Ellen) (Fraud & Investigations)

Clawback (Fraud & Investigations)

Clawback AI (Fraud & Investigations)

ClearBid (Intake & Applications)

DailyHackers (Fraud & Investigations)

Darsini Lakshmiah (Intake & Applications + LLMs vs. Gov Consulting)

DeepAudit (Fraud & Investigations)

Erinys (Fraud & Investigations)

falcon (Fraud & Investigations)

Fantasic Four (AI-powered operational intelligence for government resource deployment)

GovPulse (LLMs vs. Gov Consulting)

GrandShield (Fraud & Investigations)

Hackers (Intake & Applications)

Holocron Investigative (Fraud & Investigations)

InsomniacDesign (Intake & Applications)

Mayam Shahbaz Ali (Fraud & Investigations)

Myan (Fraud & Investigations)

NASA Exoplanet Query Interface (Modern public-facing infrastructure)

ORIN (Intake & Applications)

Provenance (Fraud & Investigations)

SaSSAI (Intake & Applications)

SBIR AI Labs (LLMs vs. Gov Consulting)

Sophylabs (Intake & Applications)

ZenIntake AI4Gov (Intake & Applications)

Photos

Drive folder with all photos

* * *