
Hesper AI wants an insurance adjuster to receive more than a machine-generated suspicion score. Its agents assemble the supporting work behind a claim investigation, with links back to the documents, records and calls that produced each flag.
The startup is focused on property-and-casualty insurance. Its launch announcement describes agents reading claim files, checking providers, seeking missing material and comparing evidence. A human adjuster retains the final decision.
That division of labor addresses a practical constraint. Recognizing that a file looks unusual is different from having time to investigate it. The investigation may involve several sources, follow-up calls and records that do not initially agree. Hesper is trying to automate that collection and comparison work rather than merely rank a queue.
Watch Hesper AI's original launch video.
The company describes specialized agents sharing findings with one another. Document analysis, provider verification and collection of missing records are presented as parts of a common workflow. The resulting evidence trail is intended to let an adjuster inspect the reason for a flag without starting the investigation again.
Hesper gives examples of discrepancies its agents have found, including an altered invoice and a provider saying that a reported visit never occurred. Those examples come from the company; they are not independently reviewed cases, and an inconsistency alone does not establish fraud.
Its current website extends the description across the claims lifecycle, from initial notice through recovery. It says people retain authority over decisions while agents gather, draft and check information. The site's performance comparisons are labeled as internal benchmarks, an important limitation when assessing claims of faster processing.
The launch reports that Hesper has processed claims with a combined value above $5 million and is working with paying production customers as well as pilots. Claim value is not company revenue, insurance savings or a measure of recovered money. The announcement does not supply the detail needed to independently evaluate those outcomes.
Co-founders Pankaj Dhariwal and Nitish Badu previously worked at Nanonets, an AI document-processing company. Dhariwal's profile credits him with building its Flow accounts-payable product and growing an engineering team from seven people to 25. Earlier, he helped establish Shipsy's container-tracking and arrival-time prediction systems. He studied electronics and economics at BITS Pilani.
Badu worked on Nanonets' go-to-market operations, including outbound sales and partnerships, and previously led go-to-market work at Super.AI. He also attended BITS Pilani. Their experience combines building document-heavy enterprise workflows with introducing AI products to business customers. At Hesper, they say conversations with claims teams repeatedly revealed the same constraint: staff knew which files deserved attention but did not have enough time to investigate them.
The product's value depends on the quality of the investigation, not simply its speed. Incorrect records, missing context or a misleading interpretation could all affect a claimant. Making the supporting evidence inspectable gives the adjuster a place to challenge an automated conclusion, but does not remove the need for professional review.
Hesper's most concrete promise is to make a fuller claim workup available to more files. The final handoff remains a person deciding what the evidence supports, with the original source close enough to check.