Soracom uses AWS to create a low-code development solution for generative AI-based multimodal IoT apps

Soracom is a leading IoT connectivity provider that uses a cellular IoT core built on AWS to connect over seven million IoT devices* to cloud environments globally. The company recently broadened its business from IoT platforms to IoT apps that control devices with data and images. As part of this effort, it released Soracom Flux in July 2024.

Soracom Flux is a low-code IoT app development tool. This solution applies rules to numeric data and images sent from IoT devices such as sensors and cameras. It then analyzes and assesses the data by combining multiple data sources and generative AI, then links it to external apps such as notification tools. Developers can easily combine actions, including event content and conditional branches, by manipulating components on a console screen.

Generative AI accepts various inputs from sensor data or camera images and enables simultaneous multimodal processing of several types of data, including text, images, audio, and video.

“Generative AI creates many possibilities,” says Motokatsu Matsui, CTO of Japan, Soracom, Inc. “For example, based on predictions derived from multimodal data across multiple data sources, it can precisely control air conditioner temperatures.”

Soracom Flux is designed to empower customers to develop IoT apps faster. As the use of IoT data progresses, so too does demand for apps that combine multiple data sources and formats, perform analysis and assessment with AI, and use the results to control devices. However, using AI requires programming knowledge and development work for AI service Application Programming Interfaces (APIs). As such, Soracom devised Soracom Flux as a service for users to develop custom solutions by simply writing prompts for generative AI in natural language.“The original concept came from an engineer at an ideathon,” says Matsui. “When we envisioned it in 2023, there were few generative AI models offering multimodal support. But with the launch of Claude 3 compatibility on Amazon Bedrock and the lightweight and inexpensive Claude 3 Haiku, we started full-scale development.”

Substantive development of Soracom Flux took just five months, with the solution launching in July 2024. AWS serverless and managed products provide the service platform, and Amazon Bedrock powers the generative AI service. Since the 2015 release of Soracom Air, a cellular IoT communication service, Soracom has run all of its IoT platform services on AWS. Its extensive lineup of serverless products and ease of system development made AWS the obvious choice for the new system. Amazon Bedrock offered a user-friendly AI service and the ability to select multiple models with the latest models always available.

Matsui explains, “As we’ve been using AWS for 10 years, we have a reference configuration that serves as a serverless architecture boilerplate. We can apply this to start prototyping immediately, which is the reason we use AWS. Given this history, Amazon Bedrock was the obvious choice: it’s lineup of essential APIs and ease in calling generative AI models through APIs were the decisive factors.”

The scalable architecture of Soracom Flux is built around AWS Lambda, Amazon Kinesis Data Streams, and Amazon DynamoDB. The solution employs microservices for scalability and faster development while also allowing loosely coupled and asynchronous data processing.

“Our focus during development was on simplicity,” says Matsui. “From the outset, we wanted to create a minimal system and then integrate it with devices, generative AI, and external systems to deliver various behaviors. With this approach, we completed development in just five months.”

Close communication with AWS representatives provided the latest Amazon Bedrock information and architectural advice.

“When recreating our prototype architecture in a production environment, guidance from the AWS service team and IoT specialists enabled us to design the optimal architecture,” says Matsui.

Soracom Flux customers are enjoying significantly faster development of IoT apps and lower operating costs. Using Amazon Bedrock for the latest versions of multiple generative AI models lowers the barriers to AI utilization and enables speedy deployment of advanced analytics such as natural language image processing.

“Historically, developing IoT apps took several months,” explains Matsui. “This includes defining requirements, connecting devices, preparing data stores, and programming to leverage generative AI. However, Soracom Flux customers can greatly shorten development times and complete these apps in days or weeks, even if it’s their first IoT app project.”

Since its launch, Soracom Flux has received immense interest from manufacturers, logistics providers, and retailers, with some users already creating applications.

According to Matsui, “Many customers want to use camera images to improve productivity for Digital Transformation (DX) and detect safety hazards, and a vast number of inquiries have involved management applications for logistics, factories, and stores.”

The Soracom Flux service is remarkably stable and free of problems at the infrastructure layer. This lightens administrator workloads and allows them to focus on the application layer.

Soracom has expanded its capabilities beyond devices and telecommunications services with Soracom Flux, which allows customers to implement their own business logic on AWS.

“Soracom Flux’s current process is initiated by event-driven architecture with sensors and images, but we’ll expand its functions for more sophisticated use of data, such as referring to previous data to reflect in the control of IoT devices,” says Matsui. “We’re looking forward to more inexpensive, user-friendly, and scalable vector data store services from AWS.”

*Connections and IoT device numbers as of November 2024.