At last week’s CES in Las Vegas, Fibocom unveiled an AI buddy to bring AI-enhanced real-time interactions to broader consumer situations.

The buddy integrates Fibocom’s self-developed AI Stack on the edge device, fuelling AI intelligence and boosting the natural interaction with end users via voice control among scenarios such as international business negotiation and global travel.

In a sleek design, the buddy acts as a personalised AI assistant that fits in the pocket. Fibocom developed it from a smart module and implemented the up-to-date ChatGPT 4o and Claude 3.5 Sonnet to the device, enabling real-time tasks such as translations, image recognition, multi-account model services, global data roaming, fast network registration and wifi hot-spot.

Additionally, by connecting to external smart wearables such as OWS (open wearable stereo) earbuds, smart glasses and smart bracelets, it forms a complete system to enable an AI-enhanced, hand-free experience via voice control, suitable for diversified scenarios.

At the core, it integrates with a Fibocom self-developed AI stack, which is a supporting tool that has a complete AI tool chain and an AI engine, accelerating the implementation of AI capabilities such as reasoning, tuning and MLLMs (multimodal large language models).

The buddy can detect and interpret voice commands instantly and improve accuracy, suitable for fast-paced mobile settings.

“As AI application builders advance AI-powered capabilities and focus deeper into the consumer-centric scenarios, the next phase of AI maturity is towards an expansion in easy-to-access, task-specific fields,” said Ralph Zhao, vice president at Fibocom. “The launch of our industry-first AI buddy truly highlights Fibocom’s comprehensive strength in the combination of smart hardware and industry-customised algorithms, and we have the confidence to manifest more future-proof intelligence to diversified scenarios.”

The AI stack is an integrated set of hardware, AI tools, AI engines and resourceful AI models that is fast to deploy for on-device AI. Laying the focus on device-side AI adoption, the stack is designed to address diverse demands for intelligent transformation across different industries by facilitating the deployment of AI models on smart devices.

Leveraging the compatibility with mainstream AI frameworks such as TensorFlow, PyTorch, Onnx and MXNet, and the support of multi-tier computing chips and modules, the stack provides a strategy for device-side AI deployment through the integration of an easy-to-deploy code and accessible AI toolchain for data annotation, model training and fine-tuning.

The stack’s inference engine ensures efficient operation, enhancing performance while reducing power consumption. This aspect is crucial for real-world applications where resources are constrained. Additionally, it supports multi-language SDK interfaces, third-party apps and a hardware accelerator.

The stack has a library of AI models for audio, CV, LLMs and MLLMs, and provides the value-added model selection and fine-tuning service to help users develop industry-oriented models. Therefore, companies of any size can benefit from the manageable service, accelerating on-device AI deployment globally.

The stack works across multiple Fibocom (www.fibocom.com) AI and smart modules, supporting the computation power from entry to premium level.

“In the wave of AI-first, we are proudly seeing Fibocom leverages AI innovations and in-depth industry expertise to empower the on-device AI deployment with our game-changer Fibocom AI stack, enabling businesses to scale smart devices in various industries with high-efficiency, cost-effective approaches,” said Willson Liu, president of Fibocom’s AI research institute. “Looking forward, we are expecting to collaborate with more industry customers to build their smart devices with Fibocom-empowered on-device AI and enable a smart future with AI productivity.”