When Patent Searching Becomes Easy

Interview with Etienne Gabella, Founder of Twenty Years and French and European Patent Attorney

Etienne, could you briefly outline your professional background?

I am an engineer and a graduate of Télécom SudParis. While at university, I co-founded a start-up developing mobile augmented-reality applications.

In 2014, I joined LLR, where I qualified as a French patent attorney and a European patent attorney. I specialised in patent protection for so-called “computer-implemented” inventions, including cryptographic systems, payment systems and, naturally, artificial intelligence.

Today, I run my own practice while continuing to work with LLR as a consultant.

How did you come to found Twenty Years?

I had long wanted to build a business in this field and make patent information more accessible.

At LLR, I conducted many prior-art and freedom-to-operate searches. These assignments are intellectually fascinating, but they can also be time-consuming and frustrating: the quality of the results often depends heavily on the keywords chosen. Once a search strategy has been defined and a frequently large number of documents retrieved, those documents still have to be analysed under tight deadlines.

I trained in AI and became convinced that semantic-similarity search was the way forward. It reduces reliance on keywords and can surface relevant patent documents in a matter of seconds. AI can also help prepare preliminary analyses across a large number of documents.

Could you explain more specifically what Twenty Years offers?

In its current form, the tool is designed principally for freedom-to-operate work. Users describe their product or invention in natural language—whether in a sentence, a paragraph or an uploaded document—and the platform returns, within seconds, the patent claims that are semantically closest to that description.

Behind the scenes, our proprietary AI models carry out several complex steps:

• identifying the relevant technical features;
• generating synthetic candidate claims;
• comparing those candidate claims, based on semantic similarity, against a corpus of more than 30 million claims from patent documents published over the past twenty years by the European Patent Office (EPO), the United States Patent and Trademark Office (USPTO), and the Canadian Intellectual Property Office (CIPO).

Based on these results, users can initiate preliminary freedom-to-operate analyses. For each document reviewed, they receive cited reasoning regarding possible literal overlap with the claimed features, potential infringement under the doctrine of equivalents, and a risk score.

Do users need to be IP professionals to use Twenty Years?

No. There is no need to master IPC or CPC classifications, Boolean queries or patent jargon. Users simply describe their product in their own words.

The tool is therefore accessible to project leads, executives, R&D engineers and researchers, providing a rapid initial overview without requiring expertise in intellectual property.

At the same time, it is particularly valuable to professionals like me, because it can substantially accelerate the early stages of our work.

Whether for patentability or freedom-to-operate purposes, confidentiality is essential. What guarantees do you offer in this respect?

For me, this is fundamental, and it is one of Twenty Years’ key differentiators.

The AI models that power the platform are entirely proprietary. They are hosted on dedicated servers provided by service providers subject to European law. No user data is transmitted to OpenAI, Google, Anthropic or any other third-party LLM provider.

This independence also helps ensure continuity of service, regardless of changes to the terms or pricing of external providers.

For LLR’s free widget, descriptions are not persistently stored. They are processed only in volatile memory and automatically purged no later than thirty minutes after processing is complete. No persistent copy is retained, and user data is never used to train the models.

How do you train the product if you do not use user-entered data?

Our models are trained on fictitious data generated in-house: technical descriptions of different lengths and styles, claims covering all technical fields and languages, and fictitious analyses supported by citations and reasoning, all validated by patent experts.

Client data never enters the training process. This is the basis of the confidentiality commitment we are able to offer.

What features are included in the free version available on the LLR website?

The widget on the LLR website enables users to:

• enter a natural-language description of up to 90,000 characters—approximately forty pages—or upload a technical PDF or text document;
• search across more than 30 million claims from patents and patent applications published over the past twenty years by the EPO, USPTO and CIPO;
• obtain the ten semantically closest patent documents, together with the title, identified applicant or proprietor, legal status, relevant jurisdictions, and the closest independent claim extracted for direct comparison.

All of this is available within seconds, free of charge and without creating an account.

What paid features does Twenty Years offer?

The extended version available on twentyyears.ai offers:

• more results per search, providing a broader overview of the patent landscape;
• search-control features, including the ability to view and modify the candidate claims generated by the tool—by reformulating, adding or removing technical features—in order to refine or redirect the results;
• a personal workspace that retains the history of users’ studies;
• PDF export of analysis reports for presentation to clients or business partners;
• access to preliminary freedom-to-operate analyses.

Will Twenty Years replace the services of a patent attorney?

Twenty Years is not designed to replace patent attorneys, but rather some of the tools they use.

In the longer term, it is also intended to help users who would not otherwise consult a patent attorney—students, founders, engineers and researchers—many of whom currently rely only on Google Patents or on general-purpose AI tools such as ChatGPT or Perplexity.

However, legal judgment grounded in human experience, discussions between inventors and patent experts, and professional opinions for which the attorney assumes responsibility all remain—and will remain—central to the role of the patent attorney.

Twenty Years is a force multiplier for patent attorneys, not a substitute for them.

What are the next developments planned for Twenty Years?

Three major projects are under way for Version 2, currently scheduled for completion by the end of summer 2026:

• geographical expansion, initially by incorporating patent publications from Japan, Korea and China, with the long-term objective of worldwide coverage aligned with the jurisdictions commonly relevant to invention protection;
• indexing not only claims, but also entire patent documents—including descriptions and examples—so that searches can cover the full content of those documents;
• a preliminary patentability-analysis module, modelled on the preliminary freedom-to-operate analysis, designed to provide an initial structured assessment of the novelty and inventive step of an invention described by the user.

Picture from Merlin Lightpainting (Pexels)