In a time when startups are racing to integrate generative AI into their platforms, Marcus Weller built something entirely new in Deepinvent, a cognitive architecture that does what no other system has done: it invents. Not repackages. Not summarizes. It creates original intellectual property, identifies underexplored areas of science and technology, and drafts patent filings with a level of nuance and depth that outpaces human inventors.
It began as a practical solution. Weller, an experienced entrepreneur with dozens of patents to his name, was helping a friend file patents for a health tech startup. Instead of manually drafting them, he designed a system that mirrored his own mental process, an AI that could spot gaps in scientific literature, analyze existing patents, and conceptualize ideas across disciplines. The outcome? A prototype of Deepinvent, capable of producing patent-worthy innovations in minutes.
Fast forward a year, and the AI had reached a point of recursive innovation. It designed and filed a patent for itself. This was more than symbolic. Deepinvent had begun operating with what Weller describes as “recursive evolutionary inference,” a brain-inspired approach where top ideas are bred and genetically diversified across cycles. In practice, that means the system learns, evolves, and predicts where innovation is heading.
At the heart of Deepinvent’s engine is what Weller calls a constellation of cognitive models. Like modular brain regions, each AI model is called upon for a specific cognitive task: scanning scientific journals, detecting commercial viability, identifying prior art, and projecting where the next frontier of innovation lies. It doesn’t rely on wrappers built on large language models. It’s a custom-built architecture that operates at the edge of what’s technologically possible.
This is where Weller’s vision diverges from the status quo. While traditional patent clusters often gravitate toward crowded, litigated zones of technology, Deepinvent is hunting for “white space,” those low-density areas untouched by existing IP. Some might argue that value lies near existing patents, but Weller sees incremental innovation as a dead end. His system, instead, explores the unmapped, combining scientific, technical, and commercial data to surface insights even experienced innovators might miss.
And it’s not just theory. One early user at Amazon’s secretive Lab126 used Deepinvent to prototype a radically different e-reader concept, an augmented reality speed-reading device that adjusts display pace based on text complexity. Within 90 minutes, he filed a provisional patent and coded the prototype. Three weeks later, he left Amazon to pursue the startup full-time.
That kind of impact, transforming casual exploration into funded ventures, is becoming a pattern. With over 2,000 inventions already generated and a waitlist of 400 companies, Deepinvent is pushing the boundaries of what startup R&D looks like. And it’s not just startups. Patent lawyers are starting to use the platform to accelerate client work. Government agencies are paying attention, too. Weller advises federal organizations like DARPA and has previously met with leadership at the USPTO to discuss how AI can help streamline the patent process, both for inventors and examiners.
The platform’s value extends beyond speed. Its real strength lies in interdisciplinary cognition. Most innovators, Weller argues, are limited by the depth of their domain. Deepinvent, on the other hand, draws parallels between phenomena across fields. It doesn’t just surface ideas; it explains them, providing users with detailed rationales and multiple invention variants to refine or combine.
This accessibility is intentional. The tool is designed to democratize innovation, with pricing tiers ranging from a free trial to a $1,000 monthly plan for startups that includes patent attorney support. Users pay with credits for core actions, like analyzing scientific literature or generating patent drafts, but interaction with the system’s new assistant feature is unlimited. That assistant acts as a translator for the AI’s deeper insights, explaining why certain technologies are relevant and helping users make sense of complex, cross-disciplinary suggestions.
Weller isn’t shy about the implications. For him, accelerating innovation is a matter of national security, not just a business goal. He sees Deepinvent as a tool to help the U.S. outpace countries like China in intellectual property generation, a process he believes underpins economic strength and geopolitical leverage.
In a landscape filled with AI hype, Deepinvent stands out not for what it imitates, but for what it originates. It’s not a wrapper. It’s not a productivity tool. It’s a generative innovation engine that sees around corners and reshapes how we invent, whether you’re a startup founder, a patent lawyer, or someone with an idea scribbled on the back of a napkin.
Want more Grit Daily Startup Show? Take a look at past articles, head over to YouTube, or listen on Apple Podcasts or Spotify.
