News

Amazon and Google Alums Launch Autonomous AI Service

• By John K. Waters
• 07/15/2020

San Francisco-based startup Abacus.AI, founded by former Amazon and Google alums, announced the general availability this week of its autonomous AI service.

The company's namesake service is designed to help organizations plug and play state-of-the-art deep learning systems into their existing customer experiences and business processes. It uses a technique its own researchers developed that finds the best neural architecture, given a particular dataset and use case. Once the neural architecture is identified, the system trains the neural net model and provides a simple prediction API that customers can use to embed predictions into their apps/Web sites. According to the company, the system takes care of all the heavy lifting, including setting up data pipelines, model re-training and providing explanations for models.

Abacus.AI (formerly RealityEngines.AI) was co-founded by Bindu Reddy (CEO), former head of AI Verticals for Amazon Web Services; Siddartha Naidu, who was a principal engineer for Amazon's fulfillment team and a developer of Google's BigQuery software; and Arvind Sundararajan, who was engineering lead for Google's ad delivery technology. Reddy and Sundararajan co-founded Post Intelligence, a startup acquired by Uber in 2017.

As part of the launch, the company is open-sourcing three novel techniques designed by its research wing to de-bias algorithms, and it's launching a model showcase feature where users of Abacus.AI services can share their models with the rest of the AI and data science community. The company has published 20 models trained on public datasets. Accuracy metrics on these models are comparable to those from models that have been hand-crafted by data scientists, the company claims

The launch drew attention this week as much for its marquee investors as its innovative technology. The company announced $13 million in Series A funding led by Index Ventures, bringing its total funding to $18.25 million. This round was led by Mike Volpi from Index Ventures, with participation from Eric Schmidt, Ram Shriram, Decibel Ventures and Jerry Yang. Following the investment, Mike Volpi and Ram Shriram will be joining the company's Board of Directors. Abacus.AI says it is going to use the funds to scale the service, gain early customers, and grow its research team.

Abacus.AI worked with 1,500 beta testers and tested hundreds of datasets to refine the accuracy of the deep-learning models created by the company's AI engine.

"Most organizations haven't been able to reap the benefits of AI/ML because it takes an army of specialized engineers and scientists to put deep learning systems into production," said Ram Shriram in a statement. "With innovative techniques like generative modeling for data augmentation and neural architecture search, the plug and play Abacus.AI service allows organizations to nimbly deploy autonomous deep learning models and instantly realize ROI."