A startup fresh out of Y Combinator just landed a $15 million Series A to tackle one of the least sexy but most expensive problems in infrastructure: compliance paperwork. Dili, led by co-founder and CEO Anand Chaturvedi, builds AI that chews through the tangled mess of federal rules — Davis-Bacon, PWA, OSHA, EPA — that govern construction projects, particularly those taking government money. The round was led by Khosla Ventures, with Allianz, Rebel Fund, Darren Bechtel of Brick and Mortar Ventures, and Y Combinator’s Garry Tan also kicking in. That brings total funding to $21.7 million.
Chaturvedi isn’t pitching another generic compliance bot. The real danger here is financial, not theoretical. “Non-compliance can result in millions of dollars of fines for those projects,” he said. The old way of staying clean — spot-checking paperwork — is a gamble. Dili’s pitch is about trading sampling for a full audit, ingesting everything from internal ERP data to vendor documents and payroll systems. The key architectural trick: large language models only handle the messy job of turning unstructured documents into structured data. A deterministic engine then sorts that data against the actual rules, walling off the risk of an LLM hallucinating a safety violation or inventing a wage rate.
The company says its software is already running on roughly 700 active projects, spanning manufacturing plants and the very data centers that power the AI boom. About half of those clients use Dili as a software tool, while the other half essentially outsource their entire compliance grind to the company on a contractor basis. It’s a hybrid model that reflects the industry’s current state, though Chaturvedi expects the pendulum to swing hard toward software. “Software and AI are going to start eating a lot of those professional services workflows,” he predicts, cutting out the middlemen who currently charge a premium to manually cross-reference government spreadsheets.
It’s a clever symmetry: the infrastructure needed to build AI is itself becoming a proving ground for AI. The question isn’t whether the technology can parse a wage determination — it clearly can. The real test is whether conservative project owners, who’ve been burned by finicky software before, will trust a system that turns a day’s work into a few minutes. A mistake here doesn’t crash an app; it triggers an audit.
