Hazel Bets That Government Procurement Is an AI Problem Worth $2.7 Trillion a Year

Government procurement is the kind of problem that makes most founders' eyes glaze over. It is regulatory, bureaucratic, and buried under decades of process. That is precisely why Hazel, a 2024-vintage Y Combinator company out of the W24 batch, finds it interesting. The company is building AI tooling to help U.S. government purchasing teams move faster, and it pegs the addressable pain at $2.7 trillion in annual public spending.

The company

Hazel was founded in 2024 by August Chen and Elton Lossner, who met the problem before they built the company. Both saw procurement dysfunction up close, Chen and Lossner citing time at Palantir and BCG where, in their telling, slow purchasing contributed to wildfire response failures in California and gaps in national security systems. The team is still small, five people at last count, headquartered in New York, and listed as active. That is roughly the size where a single product deployment can define a year.

The pitch is narrow in a useful way. Hazel is not promising to fix government. It is targeting the specific mechanics of acquisition: the lifecycle that runs from identifying a need, to soliciting bids, to evaluating vendors, to awarding and managing a contract. Its current flagship feature is AI-assisted bid evaluation, software that helps a contracting officer assess competing vendors more quickly and with more consistency than a stack of paperwork allows.

The problem they solve

Procurement is slow for structural reasons, not lazy ones. Federal acquisition rules are dense, the people doing the work are chronically overloaded, and the incentives reward caution over speed. A bid evaluation that should take days can take months. When the buyer is a school district or a defense agency, that delay has real costs, in dollars and sometimes in capability.

Hazel's wager is that a meaningful share of this work is pattern-matching and document synthesis, which is exactly what current language models are good at. Reading through hundreds of pages of vendor proposals, checking them against requirements, and surfacing the relevant comparisons is tedious for a human and tractable for a well-instrumented model. The harder part, and the reason this is not a weekend project, is doing it inside the security and compliance constraints the public sector imposes.

That constraint is the whole story behind the company's newest role. Hazel is hiring a full stack engineer specifically to deploy its platform onto classified networks, and the posting is blunt about the requirement: TS/SCI clearance with full-scope polygraph, U.S. citizens only, no exceptions. The company says it will sponsor new clearances, which is an expensive and slow commitment that few startups make unless a contract justifies it.

Traction and what it signals

Hazel has not published a funding figure, and the job listing does not name a round or a valuation. What it does disclose is more telling than a press release. The company describes this deployment as its first engagement in a new customer portfolio and says it is among the very first startups to win work with this particular agency. The role exists to bring the product into production in an air-gapped or classified environment, complete with security accreditation along the lines of IL-6 or ICD-503.

For a five-person team, winning classified federal work is a real signal. The barriers to entry, clearances, accreditation, the procurement process itself, are exactly the friction Hazel is selling against, and clearing them is its own form of traction. The company also serves the broader State, Local, and Education market, which gives it a commercial base that does not depend on a single classified contract.

The compensation tells you how the company is positioning itself in the talent market: $150K to $200K base with 0.25 percent to 0.75 percent equity, plus clearance sponsorship and relocation. That is a mid-market salary paired with the kind of equity slice that only matters if the company gets large. It is a recruiting bet aimed at engineers who want mission and ownership more than a FAANG paycheck, which is consistent with a team that calls itself "workhorses, not showhorses."

The skeptical read

Government AI is a crowded pitch right now, and plenty of companies are selling the public sector on generative models that look impressive in a demo and struggle in production. The reasonable doubt about Hazel is the same doubt that applies to any thin team taking on the most regulated buyer in the country: deploying into a classified environment is slow, the sales cycle is long, and a single agency can become a single point of failure.

The more durable case is that the moat here is not the model. Anyone can call an API. The defensible part is the unglamorous work of accreditation, security posture, and the institutional trust required to run software on a classified network at all. If Hazel actually clears that bar, the same barriers that make this hard become the thing that keeps competitors out. The stack it is building on, AWS infrastructure, Python and TypeScript, React and Next.js, PostgreSQL, Terraform for infrastructure as code, is deliberately conventional, which suggests the team understands that the hard problem is the deployment context, not the framework choices.

Whether Hazel becomes a real player or another well-intentioned govtech footnote will come down to execution on exactly the deployment this hire is meant to deliver. The procurement problem is genuinely large and genuinely neglected. The open question, as always with the public sector, is whether software can move faster than the institutions it is trying to speed up. You can follow the company and its open roles through its Y Combinator profile.