Six local startups based on Artificial Intelligence

The IDB study “Deep tech: the new wave” predicts that the majority of startups in Latin America do not aim to develop “generative” Artificial Intelligence, i.e. the type of AI that creates original content starting from existing data. But they are able to leverage innovation to create new offerings and business models.

One case is that of Teramotfounded in 2002. It is a “Software as a Service” (SaaS) platform for virtual assistants, which uses artificial intelligence to power chatbots that converse with users.

More importantly, Teramot uses artificial intelligence data engineering, which consists of understanding where the necessary information is located, how it is stored and how it must be organized to be usable. Thanks to this proposal, he relaunched $780,000 in entrepreneurial capital, mainly from CITI.

The founders of Teramot are Bruno Ruyú, Lucas Uzal and Leonardo Ruspin, who were introduced to the world of AI in 2005 at the Balseiro Institute, while studying the subject Neural Networks of the Physics Degree Course.

“Lucas suggested that I start a project, seeing the progress in image processing and reading the articles on text interpretation that were starting to be published, because it was clear that a revolution was coming,” said Ruyú, who today leads a team of 14 people.

He continues: “We identified the opportunity after experiencing the problem of data management firsthand. In all companies, managers want fast and correct information, but this is something that is difficult for them to solve. And even more, the disorder in information means that the products and services that companies offer to their customers are not the best. When we started working with GPT-type language models, we realized we could create a technology to solve this problem.”

Uzal assured that Teramot has a level of technology thanks to which 100% of questions and requests are answered correctly. “All the competitors we’ve tested are between 60 and 70%,” she said.

According to the scientist and entrepreneur, this difference obtained is due to a technology that they have in the process of patenting. “We are very clear about our main difference: “Assistants or chatbots are useless if information is not organized.”

Although the product is still in the testing phase, the entrepreneur noted that several of his users, who had started using Teramot to manage internal information, ended up using it for the commercial attention of their external customers.

Automatic interaction

Argentina has the largest number of Deep Tech startups – which offer scalable technology solutions to tackle big challenges – in the region. With a total of 103 companies which has received institutional capital and whose value reaches 1.9 billion dollars, 8% of the total corresponds to the Artificial Intelligence segment.

The Cordoban Aivo is another example: it is a platform that offers “Software as a Service” (SaaS) to automate customer service through the use of chatbots which are in turn driven by Artificial Intelligence. Aivo has BBVA, Cencosud, Movistar, Sony and Volkswagen among its customers, and was acquired last August by the American company Engageware.

“Aivo has already helped hundreds of companies automate their interactions. Now, with Engageware’s reach in the US market, we can offer our solution to thousands more. “We are confident that joining Engageware will advance our joint mission to revolutionize how organizations help their customers meet their needs on a global scale,” said its founder, Martino Frascaroli.

Not just text but voice

Even Balseiro graduates, in this case with a master’s degree in Physics, Raúl Sosa and Julián Neñer co-founded NeuralWave Technologies last May.

The target market is video game developers and multimedia content production companies, who pay by the minute generate natural voices using artificial intelligence and, in this way, reduce the costs of hiring voice actors. It is a market that, according to its data, generates a turnover of 15 billion dollars a year.

“Right now we offer voice synthesis and cloning services. In the near future we plan to add the modification of accents and emotions”, anticipated Neñer, who estimates an initial turnover of 200,000 dollars a year “after the investment of one million dollars that we are looking for”. The initiative has just been chosen as the best startup of the business accelerator University of Luxembourg.

“Our advantage over competitors like VoiceAI and Voicemod lies in the quality and low cost of our generations. Our AI models are small and efficient, which allows them to be used on any commercially available device, without too much computing power,” said Neñer.

While developing emotional text-to-speech technology for secondary languages, entrepreneurs are designing an API for use directly in other software, such as game engines. “We will include features such as translating into different languages, keeping the same voice, and changing the accent in real time. With these new applications we could target the call center market,” concluded Neñer.

Peer cooperation

Another noteworthy case, founded by a scientist, is that of egg, which has already “collected” a series of investments from more than three million dollars: $100,000 from Embarca in 2018; 825,000 from Alaya the following year; 1,000,000 from Draper Cygnus in 2021 and $1,200,000 from Air Capital in 2022.

The company managed to develop a model focused on peer learning, in order to ensure cooperative principles defined by science and with the potential to be expanded through technology.

“Each student chooses a course on Egg. During class time, you meet the challenge of the day and the classmates with whom you will solve it in a video call. People are chosen by our artificial intelligence, based on the information collected. If they can’t resolve something during the meeting, they ask for help from a mentor who, in a personalized way, joins the video call. In the process, students can recognize those who help them with impulses. As people accumulate them, they develop different roles within the platform and obtain EggCoins, which enhance their achievements and allow them to purchase other courses,” she summarized. Ignacio Gomez Portilloalso a master’s degree in Physics from Balseiro, co-founder of the initiative together with Carolina Pérez Mora and Gonzalo Innocenti.

“Since it was impossible to frame the project within CONICET, I resigned from my position as a researcher and co-founded a startup where I could channel this vision,” explained Gómez Portillo.

Fire detection

Satellites on fire It was created in 2020 by then high school students Ulises López Pacholczak, Joaquín Chamo and Franco Rodríguez Viau. The company has just made an entrepreneurial capital investment of $400,000 Of TechStars, Draper Cygnus, Air Capital and Antom, which are added to another 60,000 dollars already raised in the competitions of ITBA and the Swiss accelerator START.

Their proposal consists of a real-time early warning system for forest fires. In addition to the use of satellite images, provided by NASA’s FIRMS system together with two other satellites, it has developed an automatic fire detection model on cameras installed on the towers, which uses artificial intelligence to recognize smoke in the images.

“We were relying on a paper that showed 70% accuracy. But we have already managed to reach 78%. Furthermore, another artificial intelligence model we are developing predicts the evolution of fires based on climatological factors and the fuel load on the ground,” said Rodríguez Viau.

The service has 2,000 monthly active users and monitors 500,000 hectares in eight countries. “Our customers are forestry companies such as Empresas Verdes Argentina or fire stations, including the National Council of Volunteer Firefighters, with whom we are implementing the system in 70 operations.”

More than translation

The versatility of artificial intelligence also allows it to contribute to the localization of advertising campaigns in one hundred languages, customizing translations based on each advertiser’s brand. This is what he dedicates himself to GetGlobyFound by Diego Antista and Juan Fusoni in 2021. It has its own technology in the patenting process, called Jaga, which combines automated translation with advanced trans creation, which aims to reduce time and costs.

GetGloby has already “relaunched” 1.3 million dollars of the investment fund Newtopia and private investors. Its current clients are the international agencies Dentsu and Aleph but also direct clients such as Gucci, LATAM Airlines, McCormick, Tenkai, Alexander McQueen, Pandora and Leonardo Hotels. The business model involves a monthly subscription, instead of the cost per word charged by traditional translation agencies.

Antista, formerly of Google, said the main technical difficulty of the initiative is ensuring the quality of translations.

“Even if grammatically correct, the translations often failed to capture the essence of the brand. To address this challenge, we allowed each advertiser to complete a “brand voice,” detailing the brand voice profile, target audience, and style restrictions. Our translation engine uses that context to ensure translations are personalized,” she concluded.