He spent his PhD teaching machines to route traffic through optical networks. Then he pointed the same math at the one industry lawyers swore was AI-proof: patents.
Hundreds of billions of dollars move through the patent system every year. The software penetration was close to zero. Most founders looked at that and saw a graveyard of regulation and Latin. Chris Parsonson looked at it and saw an optimisation problem that nobody had bothered to solve.
“We haven't hired any sales or marketing people up to this point. All our growth has come organically by building the best product on the market.”
Walk into Solve Intelligence and the founding team reads like a UCL reunion. Chris is the CEO. His younger brother Angus runs engineering. Sanj Ahilan, who did his own machine learning PhD down the corridor, leads research. Three people who knew how to read a NeurIPS paper before they knew how to read a patent claim.
What they built is deceptively plain: an in-browser editor where a patent attorney drafts, prosecutes, and analyses intellectual property with an AI that has been trained on the specific, unforgiving grammar of patent law. Office action responses. Invention harvesting. Claim charts. Freedom-to-operate analysis. Invalidity challenges. The unglamorous machinery of a $100B+ profession.
The clients are not early-adopter hobbyists. DLA Piper. Siemens. Perkins Coie. Amgen. The kind of names that run procurement reviews and security audits before they let software near a client's portfolio. By late 2025 more than 400 IP teams across six continents were using it, with customers reporting efficiency and quality jumps of 60 to 90 percent.
In December 2025 the company closed a $40M Series B and rolled out “Charts,” a high-volume analysis tool that lets firms encode their own proprietary workflows as reusable AI styles. The pitch is no longer just “draft faster.” It is “run your whole IP practice on this.”
Bars are proportional to round size. The Series B alone more than tripled the Series A - and the company was profitable the whole way through.
“We've never lost in a head-to-head with a competitor. Now is the right time to scale rapidly.”
Open Chris Parsonson's academic website and you will not find a word about intellectual property. You will find graph neural networks. Reinforcement learning. NP-hard discrete optimisation. Resource allocation in optical data centre networks and the scheduling of ultra-fast optical switches.
His PhD at UCL was about teaching algorithms to make brutally hard combinatorial decisions - the kind where the number of possible answers dwarfs the atoms in the observable universe - and to make them fast. He published at AAAI, NeurIPS, ICML and the optical engineering venues JLT and OFC, picked up awards, and reviewed for the same conferences. Before that he had passed through Dyson, the Alan Turing Institute, and InstaDeep, the AI lab later acquired by BioNTech.
Drafting a patent is also a combinatorial problem. There is a near-infinite space of ways to phrase a claim, each with different legal consequences, and a vanishing subset that is both broad enough to be valuable and tight enough to survive challenge. A man who spent years compressing impossible search spaces into good-enough answers had, it turned out, been training for patents the whole time.
“Hundreds of billions of dollars are spent on patents every year. But there's very little software penetration in this space.”
“The long-term vision is to build the go-to AI platform to assist inventors, in-house teams, and outside counsel across every part of the patent process.”
The patent profession is one of the last large, lucrative, document-heavy industries that software mostly skipped. A patent application is a high-stakes literary form: too narrow and a competitor designs around it, too broad and an examiner rejects it or a court voids it. Every word carries legal weight, and the people writing those words bill by the hour. That combination - enormous spend, painstaking craft, almost no tooling - is exactly the gap Solve Intelligence walked into.
The product started where the pain was sharpest: drafting and prosecution, the back-and-forth between attorneys and patent offices. From there it grew outward into the full lifecycle. Invention harvesting, where raw engineering disclosures become filable claims. Office action responses, the rebuttals to an examiner's rejections. Claim charts and infringement analysis, the spreadsheets that decide litigation. Freedom-to-operate and validity work, the risk maps that tell a company whether it can ship.
Crucially, it was built for people who do not trust black boxes. Patent attorneys are trained to distrust anything they cannot verify, and to guard client confidentiality like a vow. So the platform leans on multi-source citations, sandboxed environments, and customisable models a firm can shape to its own house style - the kind of features that survive a procurement review at a firm like DLA Piper.
“Top 1% in focus and conviction.”
The most counterintuitive thing about Solve Intelligence is what it didn't do. For its first stretch it grew without a sales team and without a marketing team. The bet was simple and slightly arrogant: build something patent attorneys would rather use than not, and let the renewals do the talking. The revenue chart suggests the bet was correct.
Chris still keeps a public Google Scholar profile and an academic site listing his open-source benchmarking code. The founder of a fast-scaling company has not bothered to scrub the evidence that he used to be a researcher - because, by temperament, he still is one.