About

Belleau Wood. June 1918. The German Spring Offensive has punched through Allied lines and the road to Paris is open. The 4th Marine Brigade is ordered to hold the line at a wheat field outside the village. They don't just hold — they attack. For three weeks, against machine guns and artillery, through woods French command said were impossible to take. The Germans call them Teufelshunde. Devil Dogs.

The name isn't sentiment. It's a posture. Move into the gap, take the difficult ground, don't wait for someone to hand you a clear lane.

Modern Department of Defense acquisition is broken in ways that aren't getting fixed. The traditional FAR process takes two to five years. By the time a contract awards, the AI model it specified is three generations out of date.

The money, though, has quietly moved. OTA obligations hit $18 billion in FY24, up from $1.8B a decade earlier. AFWERX STRATFI runs $3-15M deals at four-year performance periods with no FAR overhead. The Defense Innovation Unit has done 450+ CSO awards with a 51% transition rate and $5.5 billion in follow-on contracts. SBIR literally lapsed for six months in 2025-26. The fast-track instruments are where AI-native startups actually have a shot.

The problem is that data is scattered across SAM.gov, DIU project pages, AFWERX press releases, and USAspending. No single place surfaces it cleanly for a VC or BD lead trying to read the market. And nobody scores AI-readiness on top of it. Govini sells to program offices. HigherGov and GovDash score contractor-fit. Nobody asks whether a given solicitation is shaped for an AI-native company, or is dressed-up IT modernization with an “AI” label stuck on.

That's the gap. Belleau Labs scores AI-readiness: five dimensions, 25 points, updated weekly.

Aaron Wilson. Marine infantry mortars squad leader, 2020-2024. Columbia sophomore studying Applied Mathematics with a minor in Artificial Intelligence.

Before the Corps I built Porsche race cars at a shop in Texas and rebuilt Trans Ams. I care more about how systems actually work than the theory of them.

While I was in I got to experiment with a lot of new products as the consumer and test them in the field. When I got out I wanted to understand the industry side. Who builds the stuff that works. Where the money actually flows. The numbers turned out to be bigger than I expected, but the data was scattered across SAM.gov, DIU project pages, AFWERX press releases, and half a dozen other places, and nobody had pulled it together or scored it for what mattered.

I started keeping spreadsheets to track it. After a few weeks they got out of hand. I built this instead.

Both repositories are public and MIT licensed. Backend is Python, SQLite, and the Anthropic SDK. Frontend is Next.js and TypeScript. Nothing behind a paywall.

The scoring rubric is explained at /methodology. Read it, poke holes in it. The v0.1 scores will look different in six months as outcome data comes in — that's the point.

Newsletter coming soon via Substack. Weekly: top-scored contracts and companies, what they mean, and what to ask.

GitHub: belleau-labs · belleau-labs-frontend