Regular updates on the latest VC-backed AI startups. Follow along to stay informed!
Fero Labs, an AI-powered manufacturing process optimization startup, raised $15M in growth funding. Climate Investment led the round, with participation from Blackhorn Ventures, Innovation Endeavors and DI Technology.
Problem to be Solved
Manufacturing plant operators lack data science skills, which makes it hard for them to unearth inefficiencies in manufacturing processes that lead to increased emissions and decreased profits.
How They Use AI
Fero Labs approaches AI in a more traditional, machine-learning sense. They are training predictive models on various tasks such as carbon emissions estimates—these are not language models. That is, instead of throwing a massive, mysterious model at a problem and blindly trusting it (ChatGPT in many instances), they train a smaller model that can be dissected to understand why it makes the predictions it does. They claim that these machine learning models can be trusted enough to significantly improve manufacturing processes.
Business Model
SaaS. The company’s revenue tripled in 2022 (off an undisclosed base, which is presumably small). Fero Labs reports that its customers have realized more than $20 million in savings while reducing carbon emissions by around 100,000 tons since using its platform.
DeepChecks
Problem to be Solved
Machine Learning (ML) systems can be fragile—especially over long periods of time where models may be subject to data drift and prediction drift (where data is slowly changing over time and therefore predictions begin to deteriorate, like a trading algorithm trained during a bull market but deployed during a bear market), so DeepChecks is building DevTools to take care of testing and monitoring ML data and models.
How They Use AI
DeepChecks builds open-source software tools that allow developers to easily track machine learning metrics and report them to an internal dashboard. The company is specifically focused on the testing, CI/CD, and monitoring steps of the software development lifecycle. ML applications have unique characteristics like performance metrics, training data statistics, etc., so developers need an ML native solution instead of a generic one from GitLab or others.
Business Model
Open-core. Developers can get started using open-source tools for free, then graduate to products with more security and architecture features and more available support for a per-model fee.
Cohere, a developer of AI foundation models, raised $270M in Series C funding. Inovia Capital led the round with participation from Nvidia, Oracle, Salesforce Ventures, DTCP, Mirae Asset, Schroders Capital, SentinelOne, Thomvest Ventures and Index Ventures.
Problem to be Solved
Cohere aims to bring large language models to all businesses. They are built on a research lab dedicated to fundamental advancements in machine learning, similar to OpenAI. Currently, the company builds proprietary large language models on transformer architecture which developers are using to solve a variety of problems in new ways.
How They Use AI
Cohere has trained many large language models that are competitive with OpenAI. Their language models can be used for search, generation, and classification. They emphasize fine-tuning (updating model parameters to teach the model new information and abilities) and are more willing than OpenAI to build custom use cases for customers.
Business Model
Usage-based API billing. Cohere bears large upfront costs to build each model, which it recoups by charging other companies for access to the models. This business model, shared by AI giants like Anthropic and OpenAI, is still unproven. The capital-intensive development investment may pay off for the few companies who lead foundation model development, but it’s also possible that the space will become commoditized if developers are able to train models for a fraction of the price in the coming years.