When you ask Stephanie Curcio how she ended up building one of the most sophisticated AI-driven patent research platforms in the world, she immediately smiles in a way that signals the improbability of her path.
“I never set out to be an entrepreneur,” she says, almost amused. “I just saw something that was going to fundamentally change the way my profession worked, and once I saw it, I couldn’t unsee it, and I felt an obligation to follow that thread as far as it would take me.”
That thread originates in Toronto, long before NLPatent existed in any form, when Curcio was studying neuroscience at the University of Toronto. Her academic life revolved around questions about brain chemistry, behavioral outcomes, and the fragile equilibrium that breaks down during neurodegenerative disease. She spent her days immersed in lab research, analyzing the cascading effects of even microscopic biochemical disruptions. It was a rigorous scientific world grounded in precision, observation, and systems thinking.
Then, in a pivot that even she describes as “somewhat random,” she applied to both medical school and law school. After speaking with her principal investigator, who recognized her strength in writing, communication, and abstract reasoning, she ended up choosing law. Once she arrived at Western University for her JD, her STEM background funneled her almost immediately into the intellectual property track. Patent law, with its intricate blend of science, language, argumentation, and strategy, fit her naturally.
She thrived. She became president of the IP club. She earned a position at one of Canada’s top IP firms. She began advising inventors, companies, and R&D teams, working on patent preparation, prosecution, and legal opinions. She worked extensively with patents, whose drafters are trained to intentionally obfuscate just enough to protect broad claims while remaining technically precise, a peculiar linguistic art form that makes patent documents some of the most complex, dense, and inaccessible texts on earth.
And then, in 2017, something happened that fundamentally reshaped her professional trajectory.
The 2017 Moment When Everything Changed
While still a junior practicing attorney, Curcio was introduced to early language-based AI systems. Before ChatGPT or consumer-facing LLMs, these were the primitive ancestors of what the world now takes for granted; small, experimental, and largely unknown outside research circles.
Yet to Curcio, they were a revelation.
“Even though the models were nothing like what we have today,” she recalls, “I immediately understood how transformative this could be for patent work. Patent documents are linguistically dense, intentionally technical, and extraordinarily difficult to search using keyword methods. The moment I saw language-based AI, it was obvious this would eventually change the profession.”
Patent search at the time relied almost entirely on keyword matching, a brittle, easily-misled process that breaks down the moment a drafter uses a non-obvious synonym, a field-specific term, or a strategically vague description. Curcio knew from lived experience how painfully slow and incomplete this process was. She also saw that the patent system itself creates the conditions for bad search results: the more complex and obfuscated the language, the more protection you can secure for your client.
She realized that if an AI system could actually understand the meaning of patent documents, not just their surface keywords, it would unlock a cascade of improvements across the entire practice: drafting, searching, analysis, strategy, competitive intelligence, licensing, and beyond.
Leaving Practice to Build the First Version
In 2018, driven by this conviction, she left the comfort and stability of BigLaw to co-found a precursor to NLPatent with a friend from law school. The original product, built on custom Word2Vec-style embeddings that attempted to capture conceptual meaning within patents, actually worked, at least on a technical level. But it was also rigid, difficult to scale, and required users to bend their workflow to the limitations of the architecture.
But by 2021, a new technological shift was emerging, and it became clear that in order to build the version of the platform she now knew was possible, a significant rebuild was necessary.
The 2021 Rebuild: Betting Early on LLMs
Re-founding the company in 2021 meant making a controversial decision at the time, to rebuild the entire platform around an LLM-driven architecture before LLMs had entered public consciousness.
“It was not at all mainstream,” she says. “But the machine learning community was clearly moving in that direction. We made an educated guess that large language models would become the foundation of semantic understanding for complex documents, including patents.”
That educated guess turned out to be prescient.
The new NLPatent platform uses a system of models, each specialized for a different function. One retrieves conceptually relevant patents. Another re-ranks them. Others analyze, summarize, identify patterns, and generate insights across datasets that would take human analysts weeks to navigate manually. The models are trained to understand the unique linguistic structures of patents, structures that confuse even advanced general-purpose LLMs.
“Patent language is intentionally obfuscated,” Curcio explains. “The correct keyword is almost never the intuitive one. Our models are designed to understand the substance, not the syntax.”
With the rebuild complete, NLPatent evolved far beyond a search engine. It became a full patent research and intelligence platform capable of handling every core task involved in understanding an invention’s prior art landscape, competitive positioning, and commercial potential.
The Adoption Curve That Surprised Everyone
Curcio assumed that large corporations, especially R&D-heavy organizations, would adopt cutting-edge tools first. After all, they have internal pressure to reduce legal expenses, compress timelines, and eliminate inefficiencies. But reality surprised her.
“Law firms are actually our most aggressive adopters,” she says. “They’re ahead of the corporations they serve. They’re willing to invest because they know their clients expect them to use the best tools available, and they want to stay competitive with other firms.”
Universities, particularly tech transfer offices, also represent a significant and enthusiastic customer segment. But the law firms stand out for how quickly they have embraced the new paradigm.
It’s a reversal of industry assumptions, a sign that the legal market, long considered slow-moving, is now changing faster than many predicted.
Her Vision for the Future of Legal Work
Curcio holds a clear, and somewhat provocative, view of where the legal profession is heading.
“In five years, the legal profession will look drastically different,” she says. “The lawyers who use the right tools will have a major competitive advantage, and many tasks that once required manual expertise will be increasingly automated. There may be fewer lawyers, or lawyers doing very different types of work.”
Still, she emphasizes that human legal judgment remains indispensable.
“AI can give you a shortcut to where your judgment needs to be applied,” she notes. “But it can’t replicate the holistic perspective a lawyer has when advising a client. Oversight is essential. Skipping too many steps is a mistake.”
It’s a nuanced view, grounded in both skepticism and optimism, the kind of perspective that comes from actually practicing law before building the tools designed to transform it.
Why NLPatent Matters
Patents shape the direction of global innovation. They influence whether technologies are protected, commercialized, challenged, or shelved. They guide billions in R&D investment. They determine competitive advantage in industries ranging from pharmaceuticals to semiconductors to consumer electronics.
Yet historically, the tools used to understand these documents have been archaic, inefficient, and fundamentally incapable of semantic comprehension. NLPatent represents a new foundation, not a layer added onto the old workflow, but a re-architecture of the workflow itself.
Curcio says it more elegantly: “Search was just the beginning. Intelligence is the destination.”
