Particularity sends AI investigators looking for junk fees and subscription traps

Particularity wants consumer-protection investigators to find suspect checkout flows before a complaint arrives. The startup is building AI agents that browse live websites, compare advertised prices with checkout totals and test whether a subscription can actually be canceled online.

The product is aimed at federal, state and city agencies. Its central promise is to turn a browsing session into evidence an investigator can examine: recorded steps, timestamps and hashes, followed by findings organized around the relevant legal elements. The company calls that output court-ready evidence. Whether it is sufficient for a particular proceeding remains a question for the agency handling it.

Founders Kavin Kumaravel and Christopher Bannon arrived at the idea through frustration with undisclosed advertising on social media. Their launch announcement asks how a regulator can understand the scale of questionable activity when its starting point is an inbox of complaints.

Watch the original launch video.

Particularity’s answer is to inspect the experience a consumer actually encounters. A website may advertise one amount and introduce a required fee later. A cancellation page may send a customer into a different process. An agent follows those steps and preserves what it saw, giving staff a record to assess rather than only a complaint’s description.

The company’s public website also describes checks for artificial urgency, such as offers presented with countdowns. These are potential signals for investigation, not automatic proof of a violation. The legal significance of a price, disclosure or cancellation process depends on facts and applicable requirements; the product’s role is to collect and organize those facts.

A second workflow begins with an incoming complaint. Particularity says it can perform a first-pass investigation within the same minute and match the complaint against evidence already held. It then maps findings to legal elements so agency staff can consider a warning or enforcement action. The speed claim comes from the company, which has not supplied an independent evaluation in the launch.

That combination puts the emphasis on investigative preparation. The agent can gather a repeatable sequence of observations, while staff still have to determine what those observations establish and what response is appropriate. Capturing the checkout path matters because the relevant evidence may be several screens away from the original advertisement.

Kumaravel previously worked at causal-inference startup Haus and on consumer protection and capital markets at the U.S. Senate Banking Committee. Bannon built security and compliance infrastructure at AWS, including work related to its EU Sovereign Cloud, according to the founders’ public biographies. Both studied at UC Berkeley.

The company is part of Y Combinator’s Fall 2026 batch. Its announcement does not name agency customers, report revenue or identify an enforcement case built using its software. It is seeking conversations with state attorneys general offices, city consumer-affairs departments and other consumer-protection staff.

The immediate test is practical: can an investigator review the captured steps, understand why the system flagged them and rely on the record when deciding what to investigate next? Particularity’s product starts with a small, concrete version of that task: follow a consumer through a website and preserve what happened.

‍