aiOla Brings Voice AI to Complex Industries – With Snowflake

In the age of AI, voice-to-text technology is prolific, serving everyone from call center agents to content creators. It’s used to automatically generate subtitles, transcribe customer interactions, route calls to the right agents and perform sentiment analysis that uncovers vital insights.

The tech has huge potential to deliver all sorts of benefits, and for those entering an increasingly saturated market, it’s important to find a niche.

aiOla is a deep tech lab for voice, speech and conversational AI. Its mission is to bring new efficiencies to enterprise customers with voice-to-data and workflow capabilities designed to serve the world’s frontline workers across industries such as aviation, logistics, shipping, manufacturing and maintenance. The company’s proprietary technology replaces the need to manually fill out forms, file reports and input information into various systems. Instead, users can do this with seamless, natural language interactions and watch forms populate themselves. 

“We’re reinventing data entry for these companies,” says Naor Fliker, Director of Product Marketing at aiOla. “Think about how much time it takes for employees to fill out all these forms. We can change the worker experience, and processes that once took hours can now be done in minutes.” 

To deliver this service, aiOla uses Snowflake to turn unstructured data into actionable insight. As a result, the company is saving thousands of dollars a month, while gaining improved speed, scalability and governance.

The AI voice market is crowded. But the majority of tools are designed for general purposes and often have trouble operating in loud places, which make them unsuitable for the engineers, aviation workers and factory floors aiOla serves.They also tend to struggle to understand industry-specific jargon, different accents and rapid switches between languages.

aiOla has addressed each of these issues to redefine industry standards. Its AI delivers unmatched accuracy and adaptability across multiple languages and in all environments. When compared to OpenAI’s Whisper, aiOla is 45% more accurate when transcribing domain-specific dialogue and reaching 95%+ accuracy levels when benchmarked against leading players on academic data sets.

To deliver this accuracy to clients, the company has to capture, structure and validate huge amounts of audio data in near-real time, and turn that data into structured tables that can feed analytics and reporting capabilities. However, the company’s legacy data infrastructure was struggling with that demand.

aiOla encountered issues with performance and cost, audio files and transcripts were stored outside of its analytics platform, and fragmented workflows were introducing errors and slowing time to insight. 

“Each of our environments was costing us more than $4,000 a month, and we had numerous environments,” says Tamir Hes, Data Engineering Team Lead at aiOla. “Enormous manual effort was required to make everything work.”

In 2024, aiOla moved its data infrastructure onto Snowflake’s AI Data Cloud with the intention of bringing new levels of speed, scalability and simplicity to its processes.

“Snowflake will store as much data as we need and gives us the compute power to work incredibly quickly,” says Hes. “Our customers want insights the moment they stop talking, so we needed a very, very fast machine. We tested other solutions, but Snowflake performed the best.” 

After migrating to Snowflake in about a month, aiOla saw benefits immediately. Not just to aiOla’s data team, but to its customers too, many of which are part of the Fortune 500.  

“When they know we work with Snowflake, they know that data is secured and organized — and that we’re using the most reliable solution available,” says Fliker. “It’s like an insurance paper for us.” 

Today, aiOla ingests all of its unstructured audio data into Snowflake, where it uses Dynamic Tables to automatically prepare that data for use. It’s then sent to its Insight Center service, where customers can configure their own reports or alerts and get what they need from that data.

Tagging is central to aiOla’s technology. It’s a process that takes place over two stages. The first is automatic speech recognition (ASR), where audio recordings are assigned various tasks and labels, and correlated with transcripts. The second is large language model (LLM) tagging, which enables aiOla’s technology to fill out custom forms with the information generated from the calls.

Its previous architecture required aiOla to build its own tagging UI from scratch with Kubernetes, as well as handle an entire management system for tagging and supervisor enrollment, and connect this to its existing data infrastructure.

Upon moving to Snowflake, aiOla was able to rebuild its tagging system on Streamlit in just three weeks, bringing new levels of simplicity to the process.

“It was very, very fast to build with Streamlit. Now we just give people a user account on Snowflake and assign them the role of tagger,” says Hes. “We went from two hours of onboarding for taggers to just five or ten minutes” — a 95% decrease.

The tagging tool in Snowflake also allows aiOla to enable encryption, ensuring customer data remains secure. The company saves its audio files in a binary format, and then uses Streamlit to revert the binary to audio again. This built-in layer of security, the speed at which the solution was built and its low costs provide aiOla with a huge advantage.

Following this initial success, aiOla plans to explore more of the platform’s features and functionality. The company intends to build a Snowflake Native App that customers can access through Snowflake Marketplace, so they too can generate insight from unstructured voice data. It also plans to move some of its data science workloads to Snowflake, and build a Data Agent with Cortex AI so users can uncover insights using natural language.

“My goal and my inspiration is to give executives at our client organizations the ability to ask direct questions of data. That way, they’re not having to explore dashboards – it can be entirely conversational,” says Hes. 

Just a year in, the company is very much at the beginning of its Snowflake journey, but the benefits so far  — including the ability to scale with demand and deliver high-speed input and output to customers — point to a promising future. One where the world’s frontline workers are free from the administrative burden of paperwork, and able to focus on what they do best.