1 month ago
9 ARTICLE AD
As we’ve reported before, enterprise CIOs are taking generative AI slow. One reason for that is AI doesn’t fit into existing software engineering workflows, because it literally doesn’t speak the same language. For instance, LLMs (aka large language models) require a lot of cajoling to deliver valid JSON.
That’s where a U.S.-based startup called Dottxt comes in, with the promise to “make AI speak computer.” The company is led by the team behind the open-source project Outlines, which helps developers get what they need from ChatGPT and other generative AI models without having to resort to crude tactics like injecting emotional blackmail into prompts (‘write the code or the kitten gets it!’).
Software libraries such as Outlines, a Python library, or Microsoft’s Guidance, or LMQL (aka Language Model Query Language) make it possible to guide LLMs in a more sophisticated way than mere prompt hacking — using an approach that’s known as structured generation (or sometimes constrained generation).
As the name suggests the focus of the technique is on the output of LLMs, more than the input. Or, in other words, it’s about telling AI models how to answer, says Dottxt CEO Rémi Louf.
The approach “makes it possible to go back to a traditional engineering workflow,” he told TechCrunch. “You refine the grammar until you get it right.”
Dottxt is aiming to build a powerful structured generation solution by being model-agnostic and offering more features — and, it says, better performance — than the open source project (Outlines) it was born out of.
Louf, a Frenchman who holds a PhD and multiple degrees, has a background in Bayesian stats — as do several other members of the Dottxt team. This grounding in probability theory likely opened their eyes to the potential of structured generation. Familiarity with IT beyond AI also played a role in their decision to build a company focused on helping others usefully tap into generative AI.
Louf’s reference to grammar is no accident. Dottxt’s is based on the premise that most of the text we interact with is heavily structured. There’s code, of course, but also many other templates that LLMs should be able to follow to actually be useful in work environments.
GPT-maker OpenAI recently introduced a form of structured generation which it calls Structured Outputs — and it gave a nod to Outlines as part of its “inspiration.”
Louf, meanwhile, sees Outlines’ popularity as a sign that there is demand for another flexible approach with more bells-and-whistles. And investors seem to agree: Dottxt has raised $11.9 million in a matter of months.
The startup pulled in a $3.2 million pre-seed round led by deep tech VC firm Elaia in 2023, followed by an $8.7 million seed led by EQT Ventures this August. In the interval, Louf and his co-founders have been focused on working to prove that their approach doesn’t impact performance. During this time demand for open source Outlines has exploded; they say it’s been downloaded more than 2.5 million times — which has encouraged them to think big.
Raising more funding made sense for another reason: Dottxt’s co-founders now knew they wanted to use the money to hire more people so they could respond to rising demand for structured generation tools. The startup’s fully remote team will reach a headcount of 17 at the end of the month, up from eight people in June, per Louf.
New staffers include two DevRel (developer relations) professionals, which reflects Dottxt’s ecosystem-building priority. “Our goal in the next 18 months is to accelerate adoption, more than the commercial side,” Louf said. Though he also said commercialization is still due to start within the next six months, with a focus on enterprise clients.
This could potentially be a risky approach if the AI hype is over by the time Dottxt seeks more funding. But the startup is convinced there’s substance behind the bubble; its hope is precisely to help enterprises unlock real value from AI.
Hugging Face CTO Julien Chaumond, who is an investor in Dottxt, recently dubbed structured generation “the future of LLMs“. So the hype is reaching down into this segment of the GenAI tech stack too.
With other tailwinds such as AI agents and the rise of smaller AI models, Dottxt’s adoption bet could pay off. “Everyone will be using structured generation in a few years, there is no doubt about that,” Louf predicted.
