Openlayer Inc., the creator of a novel platform for testing artificial intelligence and machine learning models, said today it has closed on a $4.8 million seed funding round.

The round was led by Quiet Capital and saw participation from Picus Capital, YCombinator, Hack VC, Liquid2 Ventures, Mantis VC and a host of angel investors including Instagram co-founder Mike Krieger, Instacart co-founder Max Mullen and many others.

Openlayer was created by former Apple Inc. machine learning engineers Gabriel Bayomi, Rishab Ramanathan and Vikas Nair. The startup explains that though there are numerous existing software testing platforms available to developers, most of them are designed for deterministic systems in which a given input will produce an expected output. That makes them unsuitable for the models, which are probabilistic. Until now, developers have had no reliable way to test the models, the startup says.

“Over the course of our time working across 15 different teams at Apple, we faced firsthand the same problem over and over: There is no standardized way to test and collaborate on machine learning models,” Nair explained. “Because of this, errors are often not caught until after shipping, which can result in potentially disastrous outcomes.”

Openlayer changes that, enabling teams to improve their models and datasets systematically before they’re shipped out. For instance, it can help teams to verify the integrity of training and validation datasets, and it can surface meaningful discrepancies between training, evaluation and production data. It helps to ensure models meet target performance benchmarks, and validates if they are robust enough for edge use cases by generating synthetic data to inject noise.

Other capabilities within Openlayer include guaranteeing fairness of model behavior across data subpopulations and tracking different versions of models and datasets to compare their performance. It can also explain model behavior by surfacing which aspects of a dataset were used to make a prediction.

While generative AI bots like ChatGPT are generating all of the buzz and excitement lately, there’s an urgent need for more tools on the testing, security and validation side of AI, said Constellation Research Inc. Vice President and Principal Analyst Andy Thurai. He explained that testing and validating models is a key part of any machine learning operations process, as visibility is a key consideration for any enterprise that’s deploying AI solutions.

“Most MLOps platform providers have built such a capability as part of their overall experience, but Openlayer is focused specifically on the validation component to help fast-moving enterprises and it brings some advantages,” Thurai said. “Openlayer’s verification option for verifying the integrity of training and validating datasets, and its ability to surface discrepancies between training, evaluation and production data, is both interesting and unique. Other functions, such as it model performance measurement, model tracking, dataset preservation and model explainability tools, are all standard fare that can be found on most other MLOps platforms.”

Openlayer said a number of Fortune 500 firms are using the platform to test and validate their machine learning models, uncover mistakes and diagnose why problems occur. Looking ahead, the startup plans to expand its team and improve its platform’s functionality so it can handle additional machine learning tasks. It also wants to add more sophisticated guardrails for users, so they can test their models as they iterate.

Bayomi said his vision is for Openlayer to become the go-to hub for any team building new models. “To achieve this requires building a development pipeline that delivers powerful insights about your models and data every step of the way, pre- and post-deployment,” he said.

Image: Freepik