Dili, a company building AI compliance software for federally funded construction, has raised a $15 million Series A led by Khosla Ventures. The round brings the company’s total funding to $21.7 million, according to the announcement, with participation from Allianz, Rebel Fund, Darren Bechtel’s Brick and Mortar Ventures and Y Combinator’s Garry Tan.
One clarification worth making up front, because several outlets got it muddled: the round is $15 million. The $21.7 million figure that appeared in a number of headlines is cumulative, including an earlier seed.
The Problem Is That Auditing Is a Sample
Federal infrastructure money arrives with strings. Projects drawing on federal funding or tax credits carry prevailing-wage obligations, Davis-Bacon requirements, apprenticeship ratios and certified payroll reporting — a body of rules that is intricate, regionally variable, and enforced after the fact.
The traditional way to check compliance is to pull a sample and inspect it. Dili co-founder and CEO Anand Chaturvedi puts the consequence plainly in the announcement: “Compliance in this industry has always meant hoping the sample your auditor pulled happens to be clean.”
That sentence is the whole company. If your assurance process reviews a fraction of the payroll records and the penalty regime applies to all of them, then compliance is a probabilistic bet dressed up as a control. Dili’s argument is that the economics of that bet only made sense when reading everything was impossible.
What It Actually Does
The technical shape is worth noting because it is not “point an LLM at the documents.” Per TechCrunch’s account, Dili uses AI for the extraction problem — turning unstructured project paperwork into structured data — and then applies a deterministic rules engine to test that data against the regulations.
That split matters. Prevailing-wage determination is a rules problem with correct answers, not a judgment problem, and a system that hallucinates a wage determination is worse than no system. Using models for reading and deterministic logic for deciding is the conservative architecture, and in a domain where the output has to survive a federal audit it is close to the only defensible one.
The company describes itself on its own site as an “AI-powered back office for the built world,” and says it pairs the software with expert compliance staff. TechCrunch reports roughly half of deployments run as in-house software and half as an outsourced contractor model — a services-attached shape that is common in compliance and tends to make the early revenue less clean than pure software but the retention stickier.
The Traction Numbers, and What They Are
Dili’s announcement puts numbers behind the claim: more than 700 federal projects, over $1.4 billion in wages, 5.2 million labor hours and nearly 16,000 certified payroll reports processed, and more than $50 million in fines and clawbacks caught or prevented. It also names EDF, Radiance, Heelstone and Borea among its customers.
These are company-provided figures from a funding announcement, not audited disclosures, and the customer names come from the company rather than from those firms. Treat them as claims the company is willing to make publicly, which is meaningful but not the same as verified. The $50 million in avoided penalties in particular is the kind of counterfactual that is very difficult for anyone outside the company to check.
The sector mix is the more informative detail: energy, infrastructure, construction and manufacturing. That is a precise description of where federal money is currently landing, and it explains the investor list better than the product does.
Why Khosla, and Why Now
Vinod Khosla’s stated rationale in the announcement is blunt: “One missed financial compliance requirement puts hundreds of millions of dollars at risk.”
The timing tracks a real shift. The American buildout of data centers, chip fabs and energy infrastructure has pulled an enormous volume of work into the federally funded category, and with it a compliance burden that many of the firms doing the work have not carried at this scale before. A solar developer that has always done commercial work does not necessarily have a certified-payroll function. Dili is selling to the gap between the money’s requirements and the recipient’s back office.
It also lands in a corner of construction software that has been filling in steadily. Trayd raised $10 million for specialty-trade payroll, which touches certified payroll from the processing side; Alloovium came out of stealth against construction’s compliance document sprawl; Document Crunch went after contract obligations. Dili’s slice is narrower and more regulatory than any of them — it is not trying to understand the project, only to prove the labor on it was paid correctly.
What to Watch
The bull case is that prevailing-wage compliance is a genuine wedge: mandatory, painful, penalty-backed, and adjacent to the payroll and workforce data that would let a company expand into scheduling, labor cost forecasting and workforce planning. Compliance is a good place to start precisely because nobody argues about whether they need it.
The risks are the ordinary ones for this shape of business. Regulatory products are exposed to regulatory change — a shift in federal funding conditions or enforcement posture moves the ground under the whole category. The services half of the delivery model has to become more software over time or the margins stay mediocre. And “we read 100% of the data” is a claim that competitors, including incumbent payroll and ERP vendors, can also make once the extraction problem is commoditised.
The number worth tracking is not the 700 projects. It is whether Dili’s customers start using the structured labor data it produces for anything other than staying out of trouble.
