As the demand for AI services grows rapidly, companies across industries are looking to harness its power to drive innovation and gain a competitive edge. Hailo is a Microsoft for Startups portfolio startup that provides plug-and-play conversational AI, enabling businesses to streamline processes and messaging across their most valuable platforms, services and channels, including call centers, chat solutions and SMS. increase. Their mission is to simplify digital interactions by enabling organizations to deploy conversational interfaces at scale across both text (SMS, web) and voice (call center) channels.
However, scaling and expanding AI services can be difficult, especially if you need to maintain the high level of quality that your clients expect. Hyro has recently seen an increase in demand among call centers for its adaptive communications platform, which includes automatic speech recognition (ASR) solutions and natural language understanding (NLU) capabilities, while caller frustration with speech-to-text (STT) has increased. I was also faced with evidence. ) transcription. Faced with this challenge, Hyro found a way to maintain a high level of quality by adopting key Microsoft Azure tools such as STT, Kubernetes, and eventually he Azure OpenAI Service. rice field.
Leveraging Azure services ultimately saves us weeks of human effort, helps us develop new features that differentiate our solutions in the market, and garners high interest and attention from our clients.
Exploring new AI services leads to ‘game changers’
Hyro’s AI-powered assistant has enabled customer service operations to resolve customer issues faster, but the company is facing STT transcription issues, resulting in more calls being sent to agents. was to be forwarded to This was frustrating for callers who often hung up due to poor user experience.
Concerned that these issues would damage its reputation for quality, Hyro worked to improve its conversational AI. They needed to build a speech-to-text solution that could meet the needs of clients across different industries and segments. Hyro ultimately turned to Azure services with tools such as: Azure Speech to Text, Azure Kubernetes Service, Azure blob storage, Azure Cache for Redisand Azure Open AI service.
“We conducted extensive research and evaluated quite a few different vendors along the way,” said Nitzan Bar, chief architect at Hyro. “The fact that Azure allows us to easily train a custom model for each question we ask our end-users, including yes/no questions and number-related questions, has been a real game changer for us.”
Bar praised the Azure OpenAI Service and pointed to recent use cases where the service proved invaluable. For one, Hyro needed a way to collect birth dates from users. This is harder than it sounds, considering that people around the world enter their birthdays in different orders. Using Azure OpenAI Service’s completion model, Hyro created prompts with multiple date iterations and achieved high extraction rates in just a few hours of work.
Bar said the capabilities of the Azure OpenAI Service are unmatched, and none of the competitors on the market today.
“This has brought a lot of value to the company, primarily in terms of speed of development,” says Bar. “What would take days or weeks to accomplish, now he can do in less than a day.”
Use Azure services for seamless communication
Using Azure services, Hyro can deliver high-quality conversational AI experiences that meet client-specific needs and provide users with seamless communication experiences across different platforms, services, and channels.
Azure Speech to Text: Hyro works with telephony providers to respond and resolve incoming audio data from incoming calls, using Bing’s Speech-to-Text to train the audio data with a custom voice model for transcription. The service helps Hyro transcribe audio from various sources such as microphones, audio files, and cloud storage into text in his 100+ languages. Hyro’s NLU engine can use that transcription to generate automated responses with up-to-date data. Speech models can be deployed both in the cloud or at the edge in containers and can be trained to understand organizational and industry-specific terminology.
It is now also possible to train different models depending on the different information the assistant is trying to extract from the user. For example, Hyro might have a model that is good for collecting yes/no answers to questions, and another model that collects numbers and dates. The solution is simple. Hyro only needs to provide the data (audio and tagging) and the training pipeline is abstracted away.
For more information, please visit the Azure site Azure Speech to Text.
Azure Blob Storage and Azure Cache for Redis: On startup, the NLU engine loads cached data from Azure blobs. Data and communication flows in response to user requests, using Twilio Programmable Voice to connect to Hyro to manage communications with users and record calls. When a user’s query is submitted by voice, it is sent to Azure’s Speech-to-Text service.
An anonymous, auto-generated user ID is then cached using: Azure Cache for Redis, user queries are stored in MongoDB Atlas. The communication service then uses the NLU engine as the default service to send the user’s utterances to the service that generated the answer, or if the user requests a conversation with an agent he uses Twilio Flex To do.
For more information, please visit the Azure site Azure blob storage and Azure Cache for Redis.
Azure Kubernetes Service (AKS) and Azure OpenAI Service: Hyro, a fully managed Kubernetes service with enterprise-grade security and governance, along with integrated CI/CD, leverages AKS in many ways.
- Flexible capacity provisioning with the ability to add event-driven autoscaling
- A faster end-to-end development experience
- Comprehensive authentication and authorization capabilities
After using the service before it was made available by Microsoft, Hyro began migrating to the Azure OpenAI service starting in March 2023, citing robust compliance benefits. With Azure OpenAI Service, developers will discover what is possible with OpenAI’s state-of-the-art model and leverage the enterprise-grade reliability, security, and global availability of Microsoft Azure to operate their use cases. You can bring it into the environment.
According to Bar, the company plans to launch new features based on Azure OpenAI that were previously out of scope, such as a new Q&A space that has already generated customer interest in alpha. .
For more information, please visit the Azure site Azure Kubernetes Service and Azure Open AI service.
By leveraging Azure services, Hyro was able to deliver high-quality conversational AI experiences that fit the unique needs of its clients, giving users a seamless communication experience across various platforms, services, and channels. rice field. By training speech data with a custom speech model for transcription, the company used the NLU engine to generate autoresponders using up-to-date data.
For tips on leveraging AI in your startup and other tips on accessing AI services in Azure, Sign up for the Microsoft for Startups Founders Hub today. Members receive Azure Cloud Credits that can be used for Azure OpenAI Service or OpenAI to help build products.