On March 30, 2023, natural language processing (NLP) software startup, NuMind, announced it has raised USD 3m in seed funding. The company has developed a software tool that uses large language models (LLMs) and a proprietary “Interactive AI Development” method to help its customers create machine learning (ML) models without specific in-house expertise.
The models are designed to be used for a variety of NLP tasks such as sentiment analysis, content moderation, chatbot creation, and topic detection. NuMind’s solution is available as a monthly subscription, with customers having ownership of the custom models.
NuMind was co-founded in 2022 by cousins Etienne Bernard (CEO) and Samuel Bernard (CTO). Fresh out of private beta, the company already has a dozen early customers located in the US, France, and Brazil. NuMind itself is based between Boston and Paris.
The round was led by Flybridge, a Boston-based investment firm, with the participation of big_bets, Carya Ventures, the Pioneer Fund, Velocity Ventures, Sharpstone Capital, and Y Combinator.
No Translation (Yet) But Already Multilingual
NuMind is currently focusing on text understanding (NLU) tasks and will incorporate generation (NLG) tasks — including summarization and translation — later. CEO Bernard told Slator that, although NuMind’s solution is not directly useful for transcription, it can process text after transcription.
However, since the company uses foundation models that are themselves multilingual — such as BERT and GPT-4 —, there is an inherent, multilingual element to NuMind’s solution. According to Bernard, these multilingual models render NuMind’s platform language agnostic, meaning it can “categorize data regardless of the language in which the data is provided.”
He explained, “NuMind can adapt the foundation model to the language of the users/data. When data comes in, NuMind can detect the language and choose the appropriate underlying model that has been pre-trained on this language.”
NuMind’s tool then customizes the model based on users’ data, serving as a middle layer between the foundation model and the end users. Additionally, the CEO said, “NuMind also has a version where the foundation model is multilingual (just in case the language understanding application involves many languages at once).”
Asked for examples of how multilingual models are being used by NuMind’s customers, Bernard said that there is no observable trend as yet (being that NuMind is in the early stages of customer acquisition).
However, he said, one customer is currently creating a solution to identify who in the business is most able to respond to a specific question; “Their customers typically mix questions in English and French, or English and German,” he said. Fairly typical of a company with an international / multilingual workforce.
The ultimate goal behind NuMind: to help companies that might otherwise be unable to access the capabilities of LLMs to spin up custom ML models quickly and efficiently.
Image: NuMind founders (from left) Samuel Bernard, CTO; and Etienne Bernard, CEO (Source: NuMind)