The demand for AI in the enterprise is insatiable, but the challenge lies in building the supporting infrastructure, developing and maintaining it. An IDC from 2020 questionnaire found that a shortage of data to train AI and low-quality data remain a major barrier to its implementation, along with data security, governance, performance, and latency issues. A third of companies that responded to the poll say they spend about a third of their AI lifecycle time on data integration and preparation versus actual data science efforts.
Josh Tobin, a former research scientist at OpenAI, observed the trend firsthand while teaching a deep learning course at UC Berkeley in 2019 with Vicki Cheung. He and Cheung saw the history of AI reaching a turning point: Over the past 10 years, companies have invested in AI to keep up with technology trends or help with analytics. But despite some vendors announcing the “democratization of AI,” it remained very difficult for most companies to build AI-powered products.
“The biggest challenge in building or adopting machine learning infrastructure is that the field is moving incredibly fast. For example, natural language processing was considered unattainable for industrial applications a few years ago, but is quickly becoming commonplace today,” said Tobin. “That’s why we’re building a platform for continuous improvement of machine learning.”
Tobin and Cheung, who previously led infrastructure at OpenAI and was a founding member of Duolingo, are the co-founders of portal, a service that aims to help AI development teams decide when to retrain their AI systems and what data to use during retraining. Tobin claims that Gantry, which connects to existing apps, data label services and data storage, can summarize and visualize data during the training, evaluation and implementation phases.
Gantry emerged from stealth today with $28.3 million, a combination of a Series A round of $23.9 million and a previously undisclosed seed round of $4.4 million. Amplify and Coatue led the Series A along with investors including OpenAI president and co-founder Greg Brockman and Pieter Abbeel, the co-founder of Covariant, the industrial robotics startup.
“Our product helps machine learning engineers use the data flowing through their live machine learning-driven product to find out how the application is really performing, find ways to improve it, and operationalize those improvements,” said Tobin.
AI systems learn to make predictions by recording data sets (e.g. historical weather patterns) and learning the relationships between different data points (e.g. temperature is usually higher on sunny days) within those sets. But AI systems are often vulnerable in the real world because real world data is almost never static, so the training set is not long representative of the real world. For example, an inventory forecasting system could break down because the pandemic changes shopping behavior. Volvo’s self-driving car system was infamous confused by kangaroos, because the jumping of the kangaroos made it difficult to judge how close they were.
Tobin and Cheung believe the answer to this is Gantry’s “continuous” learning system – infrastructure that can adapt a system to a continuously evolving stream of data. Gantry is designed to serve as a single source of truth for the performance of the AI system, Tobin said, allowing users to discover how the system is performing and ways to improve it using workflow tools to define metrics and the data segments on which they must be calculated.
“The days of poor corporate customer experience are over – customers now expect an experience that is as seamless, consistent and intuitive as what they have come to expect from modern technology companies. Machine learning makes it possible to deliver these experiences at scale. However, machine learning-powered products are expensive to build and pose a risk to brand and customer experiences because models can fail in unexpected and harmful ways when interacting with users,” he added. seamless machine learning customer experiences with reduced risk and cost by providing infrastructure and controls needed to securely maintain and iterate their machine learning-powered product functions.”
Gantry fits into an emerging category of software known as MLOps (machine learning operations), which aims to streamline the AI system lifecycle by automating and standardizing development workflows. Driven by the accelerated adoption of AI, analytics company Cognilytica predicts that the global market for MLOps solutions will be worth $4 billion by 2025 – up from $350 million in 2019.
Tobin acknowledges that other tools, such as Arize, Arthur, and Fiddler, accomplish some of the same things as Gantry. But he argues that they focus on a wider range of AI problems, while Gantry touches on aspects like observability, monitoring and explainability, but goes further. For example, Gantry can be used to detect bias in AI-powered apps, Tobin claims, even when the apps use “unstructured” data like text and images.
Tobin declined to reveal how many users or customers Gantry has. But he says the funding will be spent in part on customer acquisition, in addition to expanding Gantry’s 22-strong team.
“We think the potential headwinds in technology are more than offset by strong tailwinds in machine learning,” Tobin added when asked about the current economic climate and what this could mean for Gantry. “Also, as the belts tighten and companies think more about their spending, investing in tools to improve team efficiency and product performance and reliability becomes even more important.”