aiOla

The problem: information that is spoken but never recorded

aiOla was founded in Herzliya, and sources differ on when it was founded and who founded it. According to Startup Nation Central, it was founded in July 2019 by Amir Haramaty and Omri Shai. The official leadership page lists Omri Shai as founder and Amir Haramaty as co-founder. The investor New Era Capital Partners also names Guy Ernest and Erez Raanan as the company's founders, in 2019. By contrast, an interview on StartupHub.ai gives 2020 as the founding year. Ernest previously led AI architecture at Amazon. The company's starting point was field workers: maintenance technicians, factory inspectors, airport ground crews. They know what they saw and what they fixed, but reporting is done on forms and by manual typing, so much of it never reaches the organization's systems. Speech is the natural way to pass on this information, but general speech recognition fails precisely in the conditions field workers face: background noise, several speakers, heavy accents and professional terms that are not in the dictionary. "Voice is still the simplest way to get information from the field to the system," Haramaty said in 2025. "But in practice, it's still underused because so much speech recognition falls short when there's noise, multiple speakers or specialized terminology."

Jargonic: speech recognition that knows the jargon

The main product is Jargonic, an automatic speech recognition system tuned to the terminology of a specific field. Alongside it, the company developed a keyword-spotting model that, according to its own figures, detects professional terms with 95% accuracy and can be quickly adapted to new fields. The company says it supports more than a hundred languages, and claims that in noisy settings with several speakers its system outperforms those of OpenAI and ElevenLabs. This is the company's claim, not the result of an independent test. The output is not just a transcript but structured data: fields in a form, records in a system, a report that goes straight into the organization's software.

Open models

Unlike many companies in the field, aiOla has published some of its work on the Hugging Face platform. The WhisperNER model series, based on OpenAI's Whisper model, transcribes speech and, in the same pass, identifies named entities in it, such as people, places and terms, instead of running two models one after the other. In October 2025 it released Drax, a speech recognition model based on a method called discrete flow matching, along with a research paper describing it. Publishing like this lets researchers and developers check the claims for themselves, at least for the models that were released.

Funding and United Airlines

In January 2023 the company announced a $25 million raise, and at the time described itself as a developer of voice and image recognition to streamline inspection processes in traditional industry. In July 2025 it raised another $25 million in a Series A2 round, bringing its total funding to $58 million. The round included a strategic investment from United Airlines Ventures, the investment arm of the American airline, which said it was exploring the use of aiOla's tools with its ground and maintenance crews: turning spoken instructions and reports into structured data in real time.

2026: voice agents for field salespeople

By September 2026 the company had changed its focus. According to the company's website in September 2026, its mission is to give every field salesperson a personal AI agent that they talk to after a customer meeting, backed by a team of agents that update Salesforce, gather information about the customer and prepare the next visit. The same website shows new leadership: Mitch Garber as CEO and Executive Chairman, Omri Shai with the title of Founder and Executive GM, and Haramaty as Co-founder and President. The technology remained speech recognition in field conditions; the market shifted from factories and aviation to sales.