Noeri develops systems capable of progressing from business activity. The startup intends to prevent them from choosing between the cost of the most efficient models and the inferior results of cheaper options.
Choi and Guestrin bring their research experience
Noeri’s co-managing directors, Yejin Choi and Carlos Guestrin both spent several years at the University of Washington before Stanford. Guestrin contributed to XGBoost, a tool widely used in machine learning, and co-founded GraphLab in Seattle, which became Dato then Turi. After the latter’s acquisition by Apple in 2016, he led the group’s central teams dedicated to AI and machine learning.
Honored with the MacArthur Prize in 2022, Choi led research into common sense reasoning at the Allen Institute for AI (Ai2). She left the University of Washington and Ai2 in 2024 for a position as a distinguished scientist at Nvidia, held until Noeri was founded in May. She has been teaching at Stanford since the beginning of 2025. The company is based in Palo Alto and also has an office in Zurich.
Noeri suggests modifying tools like models
Noeri’s ambition for continuous improvement is particularly rooted in an experiment carried out for a Stanford team in microbiology and immunology, which sought to spend less to use OpenAI models. According to Choi, the founders designed a small model having learned to exploit older versions of these models. She says it performed better than GPT-5 for this team’s particular research needs, to the surprise of the founders.
The evening letterFree · 7 p.m.
AI transforming business?
AI strategies, movements and tricks deciphered, every evening in 5 minutes. Free.
Included upon registration: our selection of the best AI guides & comparisons.
Choose your pace
Free · No spam · Unsubscribe in 1 click
Noeri says his software looks at company data and its work processes, then suggests adjustments for a specific task. These may concern the instructions, the tools, the choice of model at each stage, or even the models themselves. The company indicates that any proposal is accompanied by an explanation and presented to a person for validation, on the principle of a code review. It announces compatibility with models with open weights as well as those with closed weights.
Founders challenge dependence on a single model
The name Noeri, which means “intelligence, in the plural”, expresses the founders’ refusal to see a single powerful model dominate uses. At the IA40 summit organized by Madrona at the end of September, Guestrin defended the idea that intelligence should not belong to one or two companies that own the models.
At the launch, Choi and Guestrin likened a centralized AI to a single provider who would also advise competitors, could freely review its conditions and could not be internalized. On property, Noeri ensures that clients keep their models and processes developed with its software. This warranty remains subject to licenses applicable to third-party models.
Menlo Ventures leads $110M funding round
Noeri raised $110 million in an operation led by Menlo Ventures, without disclosing its valuation. Seattle-based Madrona, Kindred Ventures, Wing Venture Capital and NEA are participating, along with Cisco Investments, Nvidia and Snowflake. For Madrona and NEA, it is a reunion with Guestrin: the funds had led an investment of $6.75 million in GraphLab in 2013.
Adapting models to businesses also involves other companies. Thinking Machines Lab offers Tinker, a tuning software that developers can use via API. Applied Compute, for its part, posts engineers to its customers.

