Founders Natalie Aresta-Katz and Ethan Shea built Regbase to retrieve rules, grants and government documents from poorly indexed, local and non-English sources.
By [RuntimeWire Staff](/author/runtimewire-staff)
· Published
Primary source: [Y Combinator](https://www.ycombinator.com/companies/regbase)
Why it matters #
Regbase is testing whether source discovery can become the defensible layer in legal AI, where a fluent summary has little value if the underlying rule was never retrieved. Its founders bring direct experience with both the manual research problem and the search infrastructure needed to automate it.
Natalie Aresta-Katz and Ethan Shea are building Regbase, a New York regulatory research platform that searches for laws, proposals, grants and agency guidance across countries, languages and the less accessible corners of government websites.
Founded in 2026, Regbase entered Y Combinator's Spring 2026 batch with an unusually literal version of founder-market fit. Y Combinator says Aresta-Katz spent three years as a regulatory researcher at Paul, Weiss, where she tracked legislation across roughly 45 countries. Shea built the observability backend and first Vercel Agent as a staff engineer at Vercel after engineering roles at Citadel and Microsoft. Regbase applies his search-systems experience to the research operation she once ran by hand.
Regbase is aimed at lawyers, compliance officers, researchers, regulated businesses and government vendors. In Regbase's original YC launch post, Regbase says its agents can find developments that conventional search tools and regulatory databases overlook, including draft legislation on local portals, documents published only in another language, announcements made through a regulator's social account and requirements embedded in QR-code-linked files.
The crowded legal AI market already has plenty of tools that can summarize a document once someone finds it. Regbase is betting that finding the right document early enough to act on it remains the more valuable problem.
The assignment that broke the old workflow
Aresta-Katz's path to legal technology began with soap.
Y Combinator says she sold more than $1 million of soap base on Amazon between the ages of 16 and 22 through Saponify, her online soap-making-supplies business. She was the first sustainable-business major at NYU Stern and later earned an MBA from Columbia Business School, according to the same profile.
At Paul, Weiss, Aresta-Katz spent three years tracking legislation across roughly 45 countries, relying on manual searches, local counsel and paralegals when relevant material appeared in poorly indexed or non-English sources.
Regbase's YC launch post says important updates can appear on obscure local websites, draft-bill portals, consultation pages and social media. Regbase says consultation papers might sit behind registration forms and local lawyers could still miss developments outside their usual sources.
Regbase's founding story describes a workflow that relied on manual searches, local counsel and paralegals. The example explains the economic argument Regbase is making to buyers, although the supplied materials do not establish an independently audited benchmark.
Regbase's case rests on reducing the costly first pass while preserving legal judgment: locating documents, translating them, checking their status and determining which developments deserve a lawyer's attention.
Regbase starts with discovery
Regbase divides its workflow into search, analysis and tracking. Regbase says its AI agents search original government and other online sources, including obscure government websites, draft-bill portals, consultation pages, social-media accounts and documents hidden behind QR codes or access gates. Regbase says the system matches results with official legal text and filters them according to a customer's topics and jurisdictions. Regbase's website says users can ask natural-language questions about deadlines, penalties and obligations, receive answers citing specific document sections and configure email alerts as its database changes.
Regbase advertises coverage across 195 countries and all 50 US states. Regbase's homepage also claims that the system can turn a 100-hour research project into a one-hour task. Those figures describe Regbase's stated scope and marketing claim, rather than independently verified or guaranteed performance across every jurisdiction or subject.
The unusual-source examples give the product a clearer identity. On Regbase's homepage, Regbase says it found an employment rule announced only through a minister's LinkedIn account, a compliance requirement inside a QR code on a regional ministry flyer and a government webpage that forced visitors to watch a video advertisement before revealing its contents. Y Combinator's launch post reports Regbase's claim that it found dozens of obscure laws published in non-English languages or posted only through LinkedIn, Facebook and X. Neither source provides an independent audit of those examples.
These examples expose an awkward weakness in general-purpose AI research. A model can reason over material in its context window, but it cannot retrieve a document that its search layer never finds. Regulatory research compounds the problem because relevant sources are fragmented across national, state, provincial, municipal and agency systems with inconsistent publishing practices.
Regbase's pitch therefore rests on retrieval coverage, source freshness and auditability. The quality of its prose is secondary. A polished answer pointing to an outdated rule is a compliance liability with better formatting.
The early case studies are large and self-reported
Regbase and Y Combinator describe two pieces of client work that show how the same search system can serve compliance and commercial teams.
For a protective-equipment vendor, [Regbase says it identified $400 million in expiring US school-safety grants](https://www.ycombinator.com/launches/Pxj-regbase-al-native-legal-research?ref=runtimewire). Those grants represented sales opportunities rather than legal obligations, expanding the potential buyer beyond corporate counsel to government-relations and business-development teams.
For a global manufacturer, [Regbase says it mapped 80 greenhouse-gas disclosure requirements across 45 countries](https://www.ycombinator.com/launches/Pxj-regbase-al-native-legal-research?ref=runtimewire). That project resembles the work Aresta-Katz handled at Paul, Weiss and gives Regbase a credible starting wedge: cross-border questions where manual research becomes expensive before substantive legal analysis begins.
Both case-study results come from Regbase's own materials. The cited YC launch post does not name the clients or provide an evaluation methodology. The figures consequently show the kind of work Regbase says it has undertaken, without establishing typical performance across customers or jurisdictions. Regbase does not publicly establish revenue, customer count or retention.
Enterprise adoption will depend on measurable recall, precision and update speed. Buyers also need to know how Regbase handles conflicting sources, withdrawn proposals, delayed effective dates and jurisdiction-specific definitions. Those are the cases where a missed update can turn an efficient research process into an expensive mistake.
Regbase's public position keeps humans in that loop. Regbase's website says AI should perform the first pass while people verify the results. That boundary is sensible for legal work, and it places pressure on the interface: every conclusion needs a traceable path back to the controlling source.
The incumbents are moving toward the same workflow
Regbase competes with established regulatory-intelligence providers and legal AI platforms that are expanding into continuous monitoring.
FiscalNote, a policy-intelligence platform, offers legislative and regulatory tracking from Congress down to local government, alongside policy analysis and public-affairs tools. Regology, a regulatory-intelligence platform, says it uses AI agents and a proprietary law library to monitor bills, regulations and agency updates and connect them to obligations, controls and policies.
Harvey has also published a product called Horizon Scanning, while Legora markets a legal research product.
Regbase says it targets obscure government, consultation and social-media sources that conventional search tools may not index. That positioning gives Regbase a clearer identity than another general legal chatbot, but the claimed advantage must be demonstrated customer by customer. Search coverage is difficult to measure because buyers can inspect what Regbase found while the consequential evidence may be what it missed.
Larger rivals have the resources to expand their source networks. Regbase will need to turn each obscure discovery into repeatable infrastructure: new crawlers, source mappings, translation workflows and monitoring rules that improve coverage for the next customer.
If that corpus compounds, the sources themselves can become part of Regbase's advantage. If each engagement remains a heavily customized research project, growth will look closer to a technology-enabled consultancy.
Regbase is hiring into the collection problem
Regbase has posted a product-engineer opening in New York. Aresta-Katz's regulatory experience should help Regbase distinguish a reusable research pattern from a client's special request. Shea's infrastructure background becomes important at the next stage: converting those patterns into repeatable collection and monitoring systems.
The founder's old job is Regbase's product roadmap
At Paul, Weiss, Aresta-Katz coordinated people and information across countries. Regbase replaces pieces of that research operation with software.
Shea gives that thesis technical weight. Y Combinator says he built Vercel's observability backend and first Vercel Agent. At Citadel, his group maintained the firm's master list of tradable instruments, another problem involving constantly changing records that need to remain searchable and correct.
Their backgrounds align around the same operational question: how do you maintain a dependable map of information that changes faster than people can manually check it?
Regbase has picked a useful place to answer it. Regulatory research is repetitive, expensive and consequential. It also produces clear tests. Either the system found the relevant proposal, amendment or grant before the deadline, or it did not.
The public case studies still need broader evidence, especially across customers and practice areas. Regbase nevertheless begins with a stronger foundation than an AI product searching for a legal use case after launch. Aresta-Katz already ran the workflow and worked across roughly 45 countries using manual searches, local counsel and paralegals. She and Shea are now trying to turn that operation into repeatable search infrastructure.