Abacus.AI Inc. today launched what it says is the world’s first enterprise-scale, real-time machine learning and deep learning operations platform today.

The startup’s platform enables companies large and small to stream real-time events such as clickstream data, online purchases, social media interactions and media views from websites and “internet of things” sensors. Then, that data can be processed, transformed and used to train deep learning models capable of generating contextual predictions in real-time.

MLOps is an emerging discipline in machine learning and deep learning, which are subsets of artificial intelligence. It aims to put workflows into operation by fostering more collaboration and communication between data scientists and developers.

Abacus, which has raised a total of just over $40 million in funding, reckons it can leverage MLOps to help companies of any size create powerful deep learning systems similar to the kinds of AI used by the likes of Facebook Inc., Google LLC’s YouTube and Uber Technologies Inc. Those companies have all built AI systems that leverage virtuous feedback loops in order to boost customer engagement and retention. So, for example, YouTube relies on deep learning models that can understand when a user has just watched a sports video and instantly recommend some live sports once that clip has finished.

The Abacus platform provides all of the components needed to build such an enterprise-grade AI system. It includes easy setup of data pipelines, tools for data cleaning and transformation, model training and hosting, model monitoring and explainability and even a real-time ML feature store service. Users can train their own models using Abacus’s neural network architecture, or they can specify their models on a popular framework such as PyTorch or TensorFlow and let the platform handle it.

Abacus co-founder and Chief Executive Bindu Reddy (pictured), who previously served as general manager of AI and deep learning at Amazon Web Services Inc., told SiliconANGLE that her company is bringing complex machine learning and deep learning capabilities to any company.

“Real-time deep and machine learning systems are essential for robust personalization, anomaly, search and fraud detection systems,” Reddy explained. “By bringing it to non-big-tech companies, we are making them better equipped to compete with larger companies who have more capital to build these complex large-scale systems.”

Reddy said AI systems have traditionally always required a ton of esoteric infrastructure engineering and data science talent that’s out of the reach of smaller firms. AI is incredibly complex too, with numerous moving parts that have to be managed, including streaming data ingestion, online data featurization and millisecond latency requirements for prediction, she said.

Getting a handle on all that usually requires millions of dollars of investment, putting it out of reach of many smaller players. What that means is they’re being left behind in terms of AI capabilities, she explained.

The Abacus platform changes that, eliminating the need for big investments while keeping a handle on the various moving parts involved in getting AI systems off the ground, Reddy said. She told SiliconANGLE the company’s philosophy is to “keep simple things simple, while making hard things possible.”

To that end, Abacus makes it easy to get started while also supporting more complex use cases. Customers can get begin with a simple use case such as a recommendation engine, and then expand that to a production setting simply by going through all of the steps laid out in the Abacus platform. There’s even has a low-code option that analysts can use to solve more straightforward AI problems.

On the other hand, for more skilled operators and enterprises, Abacus makes it simpler to create more advanced models by leveraging application programming interfaces, writing Python code and SQL queries. Experienced data scientists can bring their own models onto the platform.

There’s also the option to use Abacus’s AI engine to build customized deep learning models for specific use cases and datasets. In that case, Abacus’ platform will select the most appropriate neural architecture from a range of neural network types, including LSTMs, RNNs, Transformers and variational auto-encoders.

Analyst Andy Thurai of Constellation Research Inc. told SiliconANGLE that he has seen quite a few MLOps offerings emerge recently that are trying to democratize this technology. But Abacus is different, he said, in that it’s also catering to deep learning, or DLOps, as well.

“Creating a deep learning model in real-time has been a challenge for many enterprises,” Thurai said. “While the ‘digital-native’ companies have mastered it by having homegrown solutions, the legacy, hybrid companies have always struggled with this.”

The analyst said Abacus’ strategy of building on popular frameworks such as TensorFlow and PyTorch should help it gain adoption in the developer community fairly quickly. He said he likes the company’s “end-to-end AI service” approach too, which caters for DataOps for AI, ModelOps, MLOps, feature stores and AI explainability, pretty much the full gamut.

“Those are a lot of areas with different competitors that have a strong presence,” Thurai said. “If this truly works out, this can potentially become one-stop shop for all your enterprise AI needs.”

Reddy said she sees two classes of users that might adopt the Abacus platform: “There are companies that have real-time machine and deep-learning use-cases that don’t want to build all the pieces of the platform themselves,” she said. “Then there are companies with a large number of data scientists who can share features and data transformations. In this case, Abacus.AI is much simpler and easier to use than AWS SageMaker and large data science teams can go-to-market quickly.”

The CEO added that Abacus supports both supervised and unsupervised learning. It can be applied to create models for a range of use cases, including newsfeed personalization, personalized search, e-commerce recommendations, predictive maintenance in factories, cloud spend monitoring and more.

Images: Abacus.AI