There seems to be no single product or service today that is not aided by AI models and natural language processes (NLP). But implementing AI capabilities into products, and those similar to it, that can quickly process text and generate customer insights is not straightforward. The Israeli startup One AI introduces an NLP-as-a-Service platform, which offers to do just that, and it does so with ease so even people without technical knowledge or experience can use it.

Language processing as a service

The Israeli startup One AI has developed a platform that allows anyone who builds a new product to add NLP capabilities easily - without the need to write or train models. All you have to do is connect to the One AI platform and enter a text into it; the system allows you to enter continuous text or transcript of a conversation for example. Then, you can attach different ‘Skills’– each ‘Skill’ is a model developed by the Israeli startup– to produce a pipeline that you can later implement in your product.

For example, you can create a pipeline that can receive input from various texts, emit a summary, mark the speakers in the text, mark keywords, identify important topics, identify customer feelings in sales calls, anonymize text or proofreading, identify names and numbers that appear in the text and more.

One of the benefits of the One AI development is that everything happens in a very simple and intuitive ‘Drag & Drop’ interface, where all the ‘Skills’ are divided into different categories, to make it easier for you to find them and embed them in the pipeline you want to implement. Once you have reached the desired result on your sample text, all you have to do is select the development configuration of your project - Python, NodeJS, JSON or cURL - and you will receive a code snippet, which you can paste directly into your development environment, and from there to your final product.

In a conversation with Geektime, CEO and co-founder Amit Ben explains that most of the models used by One AI were built by the company's research team, along with training existing open-source models. Ben tells us that the company's models know how to recognize users' emotions - joy, anger, etc. - and not just the more basic feelings like competing companies' models’ do, such as those offered for example by cloud companies.

Divide customer service calls into the most popular topics. Source: One AI

The main challenges in integrating the language capabilities of the code, which One AI is trying to solve, are - among other things - finding the right model to perform the task; training and adapting the model to the required input, according to the industry and the specific use; and integrating the model with the other models in the product. "Our research team selects or develops the models required for each task, the team trains and adapts the models to different uses and industries. Each language task is represented by a ‘Language Skill’, which wraps several models and automatically selects the most appropriate model by input type. ‘Skills’ export output in a uniform structure, so that they can be integrated into the pipeline in order to process a language and derive different data from it in a single API call," Ben explains to us, noting the solution to all these challenges are in the product they have developed.

Ben argues that the company's significant advantage is the ability to build a useful pipeline easily and quickly for businesses - customized to the test cases relevant to it, alongside the ability to easily embed the code in the product. "Our main target audience is developers - from the individual software developer to the CTO. The secondary audience is the product managers," says Ben. According to him, products offered by technology giants like Amazon or Microsoft, through their cloud divisions, offer relatively basic capabilities - some of which are based on models that are not necessarily adapted to different tasks and types of input, so the quality of the results may be low.

Currently, like all other products dealing with the NLP world, the product works with English texts only. Ben does not rule out that the capabilities he has developed so far will also be available in Hebrew but emphasizes that it will take time for such a thing to be implemented. The company's service is offered in different subscription layers - from a basic subscription that allows the processing of one million words free of charge - and payment for transition processing.

From my experience with the One AI platform - available to everyone on the startup's site - it is a tool that has enabled me, someone who has no knowledge of programming or development, to really easily produce a product that has provided good results in less than a minute. I ran a full article on it and requested the platform to shorten it and mark dates and speakers within it - and all this came to me with the click of a button; I could have easily added more models or removed models to adapt the pipeline to different needs. From there all I had to do, if I had a product, was take the code from the platform and insert it into my product.

Investment from the founder of SentinelOne

One AI was disclosed yesterday (Monday) and announced the completion of an $8 million investment for the development of their platform from TechAviv, Ariel Maislos, SentinelOne’s CEO Tomer Weingarten, and several other investors who chose not to be mentioned. The startup was founded in 2021 by Amit Ben, Aviv Dror (CPO), Yochai Levi (CMO) and Asi Sheffer (CSO), and it employs about 20 people - half of them data scientists who specialize in NLP.

Here's how it works: