Construction / Infrastructure · Deep dive

Dili

An AI compliance platform for federally funded infrastructure — parsing certified payrolls and reading Davis-Bacon rules so contractors don't lose their IRA tax credit at audit.

emerging

The question that decides it: Do state DOLs, the IRS and Wage & Hour Division auditors treat Dili's AI-generated CPR review as evidence of compliance — enough for a contractor to defend a Davis-Bacon or IRA-PWA challenge on it alone — or does the audit-of-record stay in human hands no matter how good the model, leaving Dili to sell time savings rather than legal cover?

Snapshot

Dili is a New York YC S23 startup selling an AI compliance platform to contractors, developers and EPCs on federally funded infrastructure — clean-energy plants, chip fabs, data centres, transportation. It parses weekly certified payroll reports, apprenticeship logs and wage determinations, checking them against Davis-Bacon and the IRA’s prevailing-wage-and-apprenticeship rules. On 30 July 2026 it announced a $15M Series A led by Khosla Ventures, taking total funding to $21.7M. It says it has processed ~16,000 CPRs, $1.4B of wages and 5.2M labour hours across 700+ projects, flagging $50M+ in fines and clawbacks avoided. What makes Dili interesting: its founders originally shipped this AI to Wall Street diligence teams, then rebuilt the pipeline for the trillion-dollar infrastructure paperwork problem.

Founding story

Dili launched in mid-2023 as an AI copilot for private-market due diligence — LLM-plus-extraction to compress weeks of dataroom reading into hours. Three former Coinbase colleagues joined YC’s S23 batch: Anand Chaturvedi (Georgia Tech ML, Apple research, Coinbase, Kleiner Perkins Fellow), Brian Fernandez (Coinbase software engineer, University of Rochester) and Stephanie Song (Coinbase corp dev, ex–Galaxy Digital IB, ex-PGIM). In February 2024 Y Combinator publicised a $3.6M seed with Allianz, Rebel Fund, Lit Capital, Rocketship, Pioneer Fund and DECACORN.

Then the market told them something. Chaturvedi has described the pivot as noticing how much high-stakes audit work outside finance still ran on email and spreadsheets, with much larger enforcement dollars behind it. Federal construction spend was exploding through the IRA (~$783B), CHIPS ($280B) and IIJA ($1.2T), and every dollar came with prevailing-wage strings. In January 2025 the DOL finalised the first major Davis-Bacon overhaul in ~40 years and raised civil penalties to $13,508 per violation. Dili re-pointed its document-parsing stack at that mess; Darren Bechtel’s Brick & Mortar Ventures co-invested in the Series A, validating the pivot.

How it works

Under the hood Dili is a two-stage pipeline. First, an LLM ingests the messy unstructured inputs — scanned or PDF weekly certified payroll reports (federal form WH-347 and its state cousins), apprenticeship papers, subcontractor timesheets, DOL wage determinations — and turns each into structured records. Every worker-day becomes a row: classification, hours by type, gross pay, fringe, deductions, project number, contractor.

Second — the part the company emphasises, given the hallucination risk of AI in compliance — a deterministic rules engine runs the checks. It cross-references each worker-day against the applicable DOL wage determination for the county and craft (50,000+ active determinations, ~2M job classifications), against apprentice-to-journeyman ratios required for IRA credit eligibility, against overtime and fringe thresholds, and against state-specific overlays. When something is off — a labourer $2/hr under prevailing wage, a plumber’s apprentice above the allowed ratio, a fringe short-payment — Dili flags it against the exact CPR line and regulation. Customers fix it in-week or Dili’s team chases the sub. The platform then generates the audit-ready artefacts a DOL or IRS reviewer wants, compressing what used to be 7+ hours per organisation per week into under 5 minutes.

Product and business overview

One product — the compliance platform — sold in three flavours. Self-serve software is the traditional SaaS motion. Full-service monitoring wraps the software with Dili’s own labour: the company acts as the outsourced compliance function, chasing subcontractors and producing audit files. Roughly half of the 700+ projects run each way. A third flavour, retroactive compliance review, does a lookback for developers worried an in-progress project will fail an IRA-PWA test at credit claim.

The buyer set spans the full IRA-and-infra stack: solar and wind developers chasing the 5x tax-credit multiplier, EPCs building data centres and battery plants, GCs on IIJA transportation work. Chaturvedi has flagged the next expansion as broader audit and waste-detection work — engagements historically outsourced to the Big Four.

Business model and pricing

Dili does not publish a price list; both standard and enterprise tiers are contact-sales, per Vendr and Extruct. The evident economics: subscription fees for the software plan, plus per-project or per-CPR service fees for the outsourced flavour. Extruct AI’s public estimate pegs Dili at roughly $1.4M ARR in 2026 — plausible for 50 employees and a nascent Series A, though not a company disclosure.

The pricing power is regulatory, not competitive: on a $50M solar project a failed PWA determination can wipe out most of the 5x tax-credit uplift. That gives Dili room to price on value avoided rather than seats. The outsourced half is a hedge — if models get good enough, Dili sells more seats; if not, more managed service. Either way Dili gets paid.

Traction over time

The scale claims deserve unpacking. 700+ projects is real for this size, but “project” spans everything from a single-site solar plant to a multi-year data-centre build, and roughly half are outsourced-service engagements. The $1.4B wages figure covers processed payroll, not revenue. The $50M-plus fines and clawbacks “identified or prevented” is a customer-attributed savings number — investor-friendly, not audited by anyone outside Dili.

Market analysis

The addressable market is the compliance overhead attached to federal-money construction. Three stacked pipes drive it: the IRA (~$783B in energy and climate tax incentives, most multiplying 5x when PWA rules are met — turning compliance into the largest single line item on a solar or storage tax model), the IIJA ($1.2T authorised through Sept 2026; reauthorisation live after ~$2.3B of FY2026 rescissions), and CHIPS ($280B for semiconductor fabs). Davis-Bacon is attached to almost every dollar.

Enforcement has hardened in lockstep. In FY2025 the DOL’s Wage and Hour Division recovered $259M in back wages for nearly 177,000 workers, and the $13,508-per-violation civil penalty ceiling compounds fast on multi-year builds; a single whistleblower complaint triggers a full investigation. There is no clean top-line TAM for prevailing-wage software, but bottom-up — thousands of active EPCs, tens of thousands of IRA-eligible projects, Big Four alternatives billing at hundreds of dollars per hour — a modest software-plus-service take rate is a $1–3B category over the next five years. The caveat is that this number moves sharply with any political change to the IRA multipliers.

Competitive intel

LCPtracker is the reference point — 22 years in market, the platform state DOTs and federal awarding agencies mandate more often than any other, deep integrations, and a new investigations product (LCP CaseView, 2025). Dili’s angle of attack is that LCPtracker is a workflow tool that captures data — Dili claims to read and audit it. LCPtracker’s counter is agency-side inertia; on many jobs the contractor is told which platform to use. Elation Systems and eBacon occupy adjacent positions — Elation prime-and-agency, eBacon contractor-side with strong fringe/union and multi-state coverage. Both are older, thinner on AI. Points North’s Certified Payroll Reporting rounds out the incumbent trio.

The direct AI-native peers matter more. Reunion Infrastructure runs an auto-comparison engine against DOL wage determinations for clean-energy developers. SkillSmart InSight IQ targets the same IRA ITC/PTC unlock. DSPTCH publishes flat “TurboTax-style” pricing. Each is smaller and less funded than Dili post-Series A but attacks the same buyer with the same pitch. Behind them sit the Big Four — Deloitte, EY, PwC, KPMG — whose IRA-PWA advisory practices are the true incumbent for large audit-ready assurance work, and whose fees Dili’s Series A is explicitly designed to displace.

History and evolution

• 2023: Founded by Chaturvedi, Fernandez and Song, ex-Coinbase. Enters YC S23 as an AI due-diligence copilot for VC/PE.
• Jul 2023: Public launch as “Dili: Accelerating due diligence with AI.”
• Feb 2024: $3.6M seed announced, led by Rebel Fund with Allianz, YC principals and others.
• 2024–2025: Quiet pivot to construction and infrastructure compliance; total seed capital reportedly ~$6.7M with follow-ons; Chaturvedi steps into the sole CEO seat.
• Jan 2025: DOL rolls out overhauled Davis-Bacon regulations and a new WH-347 form; penalty cap raised to $13,508 per violation.
• 2025–2026: Scales to 700+ projects; publishes the $1.4B / 16k-CPR / 5.2M-labour-hour numbers.
• 30 Jul 2026: $15M Series A led by Khosla Ventures, with Brick & Mortar, Allianz, Rebel Fund and Garry Tan; total $21.7M.

The stumble worth naming is the pivot itself: PE-VC diligence to Davis-Bacon compliance in under two years is a hard right turn, and the seed investors bought a different company than the Series A investors did. That Khosla and Brick & Mortar are in the same round is evidence the second pitch is more investable — but it is also, cleanly, a rebuild.

What people say

The case for. TechCrunch and GlobeNewswire coverage of the Series A leans on operational metrics — review cut from 7+ hours/week to under 5 minutes per organisation, $6M in IRS penalty exposure caught for one customer, $50M+ aggregate fines and clawbacks avoided. Khosla and Bechtel frame the bet as putting AI where regulation is unforgiving and incumbents are legacy software. The YC network doubling down (Tan, Allianz, Rebel Fund) is a strong internal signal; Brick & Mortar plugs Dili into the built-world buyer network.

The complaints. Three critiques recur. First, AI hallucination risk in a compliance product: trade publications on AI-in-audit have documented LLMs inventing certifications and rules, and a false-negative on a Davis-Bacon flag becomes a lawsuit. Dili’s deterministic rules-engine stage is a real answer, but the extraction step is still an LLM, and no auditor of record — DOL, IRS, or state DOL — has publicly said it will accept AI review as evidence. Second, incumbent capture: LCPtracker is mandated on a large share of federal jobs, so Dili often ends up as an internal tool run alongside the required platform — capping ARPU. Third, regulatory dependency: the IRA-PWA multiplier is the entire reason clean-energy developers buy this software. A future administration diluting the multiplier, or IIJA reauthorisation stalling after its Sept 2026 sunset, hits the market immediately. And LCPtracker’s own knowledge base acknowledges the tool has many features contractors do not use — code for “the UX is hard” that contractors on Reddit echo. If AI-first tools like Dili do not fix that at the mandated-platform layer, the pain persists no matter how good the model.

Outlook: the open question

Dili’s Series A resolves on whether federal and state auditors ever treat AI-generated CPR review as evidence, not just as workflow. For it to break Dili’s way, three things must be true. First, the DOL, IRS and state Wage & Hour divisions accept — even implicitly, via non-enforcement of AI-reviewed audits — that a Dili-produced audit file is defensible. Second, Dili converts the agency-mandate ceiling by getting awarding agencies to specify it by name on marquee IRA and IIJA jobs, not just tolerate it as an internal tool. Third, the enforcement tailwind holds — the IRA-PWA multiplier survives the next political cycle and IIJA gets reauthorised after its Sept 2026 sunset with prevailing-wage strings still attached.

For it to break the other way is easy to imagine. Auditors keep treating AI output as advisory, so the contractor still pays a human reviewer at the end and Dili sells time savings rather than legal cover. LCPtracker’s agency mandate holds, capping the SaaS ceiling. Reunion, SkillSmart and DSPTCH — plus whatever the Big Four ship next — chew up the same buyer set. The political base case for the IRA multiplier surviving the next Congress unchanged is not obvious. The $21.7M proves the founders’ second bet; it is not enough to prove the third and deeper one — whether AI can own regulated assurance. The next raise, and whether it is priced on services revenue or a real software multiple, will say more than any press release.

Sources and further reading

• TechCrunch — “Dili raises $21.7M to bring AI compliance to the infrastructure boom” (30 Jul 2026)
• GlobeNewswire — “Dili Raises $21.7m from Khosla Ventures to Bring AI-Powered Assurance to America’s Infrastructure Boom” (30 Jul 2026)
• FinSMEs — “Dili Raises $15M in Series A Funding” (Jul 2026)
• Y Combinator — company page for Dili (S23): “Automating Compliance for Construction”
• Y Combinator launch — “Dili: Accelerating due diligence with AI” (Jul 2023)
• Points North — “The True Cost of Davis-Bacon Violations” (2025, $13,508/violation)
• U.S. Department of Labor — “Prevailing Wage and the Inflation Reduction Act”
• IRS — “FAQs about the prevailing wage and apprenticeship under the Inflation Reduction Act”
• Funding Landscape — “Infrastructure and IIJA Funding in 2026”
• Reunion Infrastructure — “Prevailing Wage & Apprenticeship (PWA) Compliance”
• LCPtracker — “LCP CaseView” investigations product launch (2025)