Credal AI Inc. and Datasaur Inc., two venture-backed artificial intelligence startups, today marked milestones in their respective efforts to ease enterprise machine learning projects.

Many enterprises are opting to build custom AI modes instead of using commercial, cloud-hosted neural networks. Often, the decision is motivated by concerns about commercial neural networks’ security. Credal and Datasaur both focus on making it simpler for enterprises to build secure AI software. 

Credal today announced that it has raised $4.8 million in seed funding for its namesake platform, which helps companies protect the input data they enter into AI models. Datasaur, in turn, debuted a new tool called the LLM Lab. A company building custom AI models for cybersecurity reasons, or simply because its requirements aren’t met by off-the-shelf models, can use LLM Lab to speed up development.

Securing AI input

When an enterprise uploads business records to a third-party AI model for processing, there’s a risk of data leaks. The company that operates the AI model might experience a network breach, or a configuration issue may cause the neural network to accidently leak the uploaded information. Credal has built a software platform that promises to mitigate such risks.

Using the company’s platform, organizations can remove sensitive data such as credit card numbers from a record before uploading it to an AI model. For added measure, Credal creates a log that tracks what information is uploaded to neural networks and by whom. This log helps companies determine if internal business data is used in breach of their AI security policies.

According to Credal, its platform also simplifies a number of related tasks. If an application uses multiple AI models, the platform can route each user request to the model best suited to carry it out. Credal enables companies to track how the neural networks that power an application are used as well as measure the associated costs. 

TechCrunch reported that Credal currently has 11 customers. The company will use the $4.8 million seed round it announced today to expand its installed base, hire more employees and build new features. The investment was led by Spark Capital.

Speeding up AI development

Datasaur is a San Francisco-based startup backed by Y Combinator, OpenAI LP president Greg Brockman and other investors. It provides a software platform that aims to ease the annotation phase of AI projects. That’s the process of enriching training datasets with contextual information to help AI models understand them.

Against the backdrop of Credal’s funding news today, Datasaur introduced a new tool called LLM Lab. The tool aims to provide an “extensive starting point” for companies that are building custom AI models because their cybersecurity requirements aren’t met by commercial, cloud-hosted neural networks. Datasaur says LLM Lab can be used to train custom neural networks on organizations’ internal data.

According to the company, the tool also streamlines other AI development tasks. LLM Lab can help software teams find ways of reducing the infrastructure costs associated with running an AI model. Additionally, the tool makes it possible to compare different foundation models and find the one that best meets a project’s requirements.

LLM Lab will support foundation models such as Llama 2, an open-source language model that Meta Platforms Inc. released earlier this year. The tool will also work with a number of supporting technologies that are often used in AI development projects. Those technologies include Pinecone, a vector database designed to store neural networks’ data. 

Image: Unsplash