zapp2photo - stock.adobe.com
As AI development accelerates into 2021, AI startups around the world number in the hundreds, if not thousands. As AI grows quickly, startups are adding advanced technologies into their products and services.
New and existing law firms, financial institutions, retailers, technology vendors and manufacturers -- nearly all larger enterprises in most industries are using or selling some form of AI, whether it is natural language processing (NLP), machine learning, deep learning or advanced analytics.
The following group of vendors to watch in 2021 comprises only a small fraction of AI startups. But these vendors appear likely to disrupt AI, either with innovative technology or easy-to-use and simple AI-based products.
While many other startups are worthy of tracking, these vendors show exceptional promise.
Law and innovation
Numerous technology vendors and law offices sell and use AI-powered legal software aimed at easing the work of lawyers and paralegals. Among other things, these tools can spot errors in legal documents or quickly create contracts.
One such startup, FairShake, is built on a unique business model.
Based in Oakland, Calif., the three-year-old vendor offers consumers a simple method to submit complaints and initiate legal actions against more than 50 preselected companies.
Many of the prelisted companies are financial or telecommunications corporations, as consumers typically file claims in these industries, said Teel Lidow, founder and CEO of FairShake.
FairShake tends to see many claims in which a corporation isn't honoring a promotion or used misleading sales tactics, Lidow said.
On the FairShake website, users can file a complaint in plain language, listing details about their problem, when it started and how it happened. Behind the scenes, machine learning and NLP models help turn users' natural language into legal documents that FairShake automatically files with the American Arbitration Association.
Users can sign up for the service for free. FairShake charges a commission if they are awarded a refund or cash settlement.
Typically, document generation and filing are difficult processes, Lidow said. "Your average consumer would be unable to grapple with it themselves," he said.
FairShake makes it simple. The company has processed some 10,000 legal claims since it started.
Another startup in this niche, Hypergiant Industries, purposely tries to look futuristic, featuring science fiction-style art on its website, the tagline "Tomorrowing today" and space age solutions and galactic systems divisions.
The futuristic look, can at first seem disorienting, but it makes sense after digging into the work Hypergiant does.
Boasting major clients such as NASA, the Department of Homeland Security, Shell, GE Power, Booz Allen Hamilton and the U.S. Air Force, Hypergiant develops AI-based products that, at one time, would have seemed fantastical.
"We formed Hypergiant to effectively tackle AI and emerging tech," CEO Ben Lamm said.
Ben LammCEO, Hypergiant Industries
Hypergiant's R&D Labs is working on an astronaut helmet that shows information, such as the wearer's vital signs, on the helmet display. Another helmet Hypergiant is designing, created for people on the ground, aims to use augmented reality to give users a video game-like map on their display, letting them see objective points and the locations of team members in real time.
The vendor is also developing a robot that can clean a room autonomously with germ-killing UV light. With optical and laser imaging, the robot could identify when humans are nearby and turn off the light, which could potentially harm a person.
"We try to be innovative around solutions we think the world needs," Lamm said, adding that, for the most part, it's investing in the R&D products it has talked about publicly "for the long haul."
While space age-sounding, Hypergiant's products and services are grounded in real technology, such as natural language processing, computer vision, robotic process automation and deep learning.
The startup, formed in 2018, also has a proven track record with more standard projects, having built a variety of applications for its customers, including a predictive fueling system for Shell customers, a rail fleet management system and an application for automating shipment documents.
Pure machine learning
Tecton.ai, created by a team that worked on Uber Michelangelo, Uber's machine learning platform, came out of stealth in April 2020 intending to make it easier for data scientists to build production-level machine learning models.
The startup sells a feature store, a repository to store and manage machine learning features and run data pipelines to transform raw data into feature values. The feature store enables users to put their features more easily into production at scale, keep track of the versions and lineage of their features and monitor the health of their feature pipelines, according to Tecton.
Tecton released the feature store as a limited beta in April and made it generally available in early December.
As Tecton puts it, feature stores run data pipelines that turn raw data into feature values, while also storing and managing feature data. With Tecton's feature store, users can more easily productionize new features; track feature versions, lineage, and metadata; and monitor the health of their feature pipelines.
The startup has raised $25 million in seed and Series A financing and $35 million Series B financing co-led by Andreessen Horowitz and Sequoia Capital.
"We think that the Tecton team is so strong that we really focused on them from an early time on," said Martin Casado, general partner at Andreessen Horowitz and a Tecton board member.
Training and document data
Founded in 2015 and backed by Amazon, DefinedCrowd sells training data for building NLP engines, intelligent speech models, and image and video recognition models.
Noting that he hears a lot about the challenges enterprises face finding high-quality training data, CCS Insight analyst Nick McQuire noted that companies like DefinedCrowd "help with the ability for customers to crowdsource high-quality training data for a range of applications."
For training speech models, the startup can source, transcribe, correct and validate speech data. For training text-based NLP models, it can collect text data, annotate, validate and tag it. Similarly, the vendor can collect, identify and validate images and videos for those models.
DefinedCrowd completed a Series B funding round earlier this year, raising $50.5 million. Previously, the startup raised $1.1 million in seed funding, and more than $11 million in a Series A funding round.
Meanwhile, yet another new AI vendor, Alkymi, emerged from stealth at the beginning of 2020 with $5 million in seed funding and a product, Alkymi Data Inbox, that automatically extracts data from business emails and documents.
Alkymi Data Inbox "tackles the problem of complex data and content within email inboxes," a big challenge as email boxes are typically locked silos, said Alan Pelz-Sharpe, founder and principal analyst at Deep Analysis.
The product brings users' emails and documents, including images, text-based files and PDFs, together into a single platform.
Using NLP and image recognition and occasional input from users themselves, the platform extracts important data and places it into appropriate tables. For example, the platform can pull payment information, including the amount and date due, from an email or document discussing billing.
Alkymi has built its platform for the financial industry, but the startup has plans to branch out to other sectors in the future.