Dottxt’s Breakthrough in Structured Generation

author
By Tanu Chahal

19/10/2024

cover image for the blog

In the world of software engineering, adopting generative AI models can seem daunting. One of the primary concerns is ensuring that these models fit seamlessly into existing workflows, rather than becoming a roadblock. The art of structured generation has emerged as a game-changer in this regard, enabling developers to harness the full potential of large language models (LLMs) and generative AI.

For enterprise CIOs, adopting generative AI has been a gradual process. LLMs, in particular, have been a challenge due to their requirement for sophisticated prompts to produce valid JSON outputs. However, the advent of structured generation has made it possible to guide these models in a more precise manner.

Dottxt, a U.S.-based startup, is pioneering the structured generation space with its innovative approach. The company's mission is to empower developers to tap into the immense capabilities of generative AI, bridging the gap between AI and software engineering workflows. By making AI 'speak computer,' Dottxt enables developers to refine grammar and output to achieve desired results.

The Outlines open-source project, developed by the same team behind Dottxt, paves the way for structured generation. This Python library allows developers to extract valuable insights from ChatGPT and other generative AI models without resorting to unconventional methods. Dottxt builds upon Outlines, offering a more comprehensive solution with increased features and performance.

At its core, structured generation is about focusing on the output of LLMs rather than the input. By refining grammar and output, developers can achieve precise results. This approach aligns with the principles of probability theory, as demonstrated by Dottxt's CEO, Rémi Louf, who holds a PhD in Bayesian stats.

The growing demand for structured generation tools is evident, with over 2.5 million downloads of the Outlines open-source library. Dottxt's $11.9 million funding round, led by Elaia and EQT Ventures, reaffirms the market's enthusiasm for this innovative solution.

Dottxt's co-founders recognize the importance of building a robust ecosystem around their product. The startup is expanding its fully remote team to 17 members by the end of the month, with a focus on developer relations and commercialization. While the AI hype cycle may eventually subside, Dottxt is confident that its structured generation approach will remain a vital part of the GenAI tech stack.