IPVectors launches a §103 engine that searches the science outside the examiner’s classification codes, then makes four rival AI models argue the result in front of a human who has to sign it. The Word add-in that audits the answer is on the Microsoft Marketplace today.
Seattle · November 5, 2026
SEATTLE — November 5, 2026 — Every prior-art report ever delivered has asked the client to take the analyst’s word for it. IPVectors today launched a patent analysis platform designed to end that arrangement.
The two-part system pairs IPV Reporter, a browser-based environment where the analysis is done, with IPVerifier, a Microsoft Word add-in that audits a finished report inside the document itself — checking each statement against the source it cites, grading it, and jumping the reader to the exact passage with one click. IPVerifier is available today on the Microsoft Marketplace.
Behind both sits the problem the platform was built around. An examiner searches the classification codes assigned to an application. The conference paper, clinical trial record or dissertation that would have made the invention obvious sits in none of them. The patent issues. Years later, in litigation or diligence, somebody finds the paper.
IPVectors maps claim subject matter into the taxonomies scientists use for their own literature — IEEE, SPIE, ACS, MeSH and Scopus ASJC among them — each held against a maintained concordance to the CPC classification, so a search can start at the claim, cross into the scientific vocabulary and walk the citation graph outward.
Every citation returned is resolved against the real bibliographic record and comes back verified, ambiguous or rejected.
“Everyone else is asking whether you can trust the AI,” said Jon Karlin, founder and CEO of IPVectors. “We built a process where every statement is checked against its source — so it does not matter whether you trust the AI, because you never have to.”
The §103 Gap-Hunter workflow cuts a claim into its limitations and opens one ledger row per limitation, each carrying its evidence and the name of who closed it. The engine scores. It does not decide.
The company’s published benchmark runs the platform against US 9,730,910 B2, a patent a United States court held obvious in 2022 — chosen because the references the court relied on are public record. Working from the patent number alone, with the search gated to art available before the priority date, the engine surfaced both of the references the court relied on.
The analysis is then argued by the Adversarial LLM Council — four models from four competing vendors in fixed legal roles: xAI’s Grok as Plaintiff, Anthropic’s Claude as Defender, Google’s Gemini as Judge, OpenAI’s GPT as Jury. In a demonstration released by the company, the council audited IPVectors’ own prior-art report and flagged one of its own statements as only partly supported, with one of the four seats scoring it red.
“We published the screenshot of our system flagging our own work, because that is the only proof worth anything here,” Karlin said. “Anybody can demo an AI agreeing with itself.”
A firm can hand a client both the report and the add-in, so the client runs the audit in their own copy of Word, with their own attestation on the record.
“Prior art does not have a side,” Karlin said. “We find the record. The legal judgment belongs to the professional, and the file shows that it did.”
IPVerifier is available now on the Microsoft Marketplace. IPV Reporter is in controlled release. The platform is covered by seven filed provisional patent applications. A user guide, sample reports and a recorded walkthrough are at ipvectors.com/guide.
IPVectors builds patent research and verification tools for professionals who have to defend the answer. Founded by Jon Karlin, sole inventor on its patent-pending technology. www.ipvectors.com
Media contact — Jon Karlin, IPVectors · jonkarlin@ipvectors.com
IPVectors produces screening work product. It does not provide legal advice. Product and company names mentioned are the trademarks of their respective owners; their use describes system components and implies no endorsement or affiliation.