Despite increased adoption of LLMs in entreprises, AI models still cannot be easily used in production environments. They are unable to generate text that respects both human-machine and machine-machine language rules, and thus they require human intervention.
We are excited to partner with Dottxt, an ecosystem and platform for interacting with large language models (LLMs), designed to deliver the most accurate and high-performing structured language generation available. Using dottxt, LLMs can output language with vastly enhanced usability through customized, structured generation.
Co-founders Rémi Louf, Dan Gerlanc, and Brandon Willard have worked together in various capacities for the past four years. During their last stint at New York-based AI firm Normal Computing, they started encountering issues with GPT-4, specifically extracting data and information in a systemized order to bypass vast amounts of manual work. While solving this issue the team invented an ingenious solution rooted in their unusual background in statistical modelling and compiler technology.
A year later, their open source library, Outlines has received over 3 million downloads, 600,000 of which were in the last month. Companies, large and small, are using it in production. Artificial Intelligence giants such as OpenAI and Cohere are among the early users of the platform.
dottxt’s mission is simple: make AI programmable so its potential meets the reliability demands of real-world systems.
By allowing the user to request information with a specific structure, dottxt transforms LLMs into tools that integrate seamlessly with existing digital ecosystems. dottxt’s functionality elevates LLMs from simple back-and-forth chat functions to dependable computers:
- Data scientists can make natural language queries to a database with a guarantee that the query will succeed;
- An HR manager looking to hire and facing a mountain of CVs can ask the service to filter through looking for specific experience or qualifications, saving significant time;
- A designer looking to extract specific attributes from a large bank of images can send the service the images and list of attributes and get the information back quickly.
Rémi Louf, CEO at dottxt, highlights:
“Everyone will be using structured generation in a few years, there is no doubt about that. Model providers, including OpenAI, are lagging in terms of speed and capabilities, and we’re here to fill that gap. With these funds, we will keep pushing the limits of this technology and make it more widely available for everyone. We are shaping a future where generative AI delivers on the kind of automation we were promised.”
On why we partnered with dottxt, our Partner Sia Houchangnia, comments:
“LLMs already have the potential to unlock tremendous value, but their lack of reliability has been the key barrier to wider adoption. dottxt solves this problem. By applying their expertise in Bayesian statistics, Remi, Brandon, Dan, and their team have created the most accurate and high-performing platform for structured generation. We are proud to be day-one backers of the company and are convinced that dottxt has the potential to become the default framework for LLM programming.”
We believed in the team from day one, writing a cheque in their $3.2 million Pre-Seed round, completed in December 2023 and led by Elaia. Dottxt’s fast growth attracted investors’ attention and the team closed an $8.7 million Seed round led by EQT Ventures, in August 2024 in which we also participated. Common Magic, Kima, FSJ, Roxanne Varza (Station F), Erik Bernhardsson (Modal Labs, CEO), Julien Chaumond (Hugging Face, CTO), Bob van Luijt (Weaviate, CEO) and Jean-Louis Queguiner (Gladia, CEO) also contributed to the rounds.
The funds will be used to expand the engineering team and bring in a Chief of Staff.
For more information, visit dottxt.co.