How RiseLink is Giving Every Object a Brain

Hermes welcomes Diana Zhu from RiseLink. Diana shares the company's 20-year history, starting with founder Dr. Pengfei Zhang's pioneering work on Wi-Fi dual bandwidth chips, leading to the founding of Beacon Corporation with 1,200+ patents, and the recent launch of the US office to make ultra-low-power edge AI chips accessible to American builders.

Diana shares her path from Applied Math at Harvard to healthcare economics at Yale, developing AI algorithms for patient-provider matching and drug effectiveness. She explains how seeing the BK7258 devboard sparked her childhood dream of making toys talk, leading her to transition from health tech to leading RiseLink's US office.

Discussion on RiseLink's mission to support US developers with both hardware and software, addressing the challenges of time-to-market and the complexity of hardware development beyond software. Diana emphasizes full engineering support to help developers turn cool ideas into reality.

Deep dive into RiseLink's world-record ultra-low-power consumption achievement of 50 milliamps in Wi-Fi keep-alive mode. Diana explains the aggressive power-saving techniques including custom wireless radio, intelligent power gating to unused peripherals, and firmware-level scheduling algorithms.

Diana explains how the BK7259 chip achieves the difficult balance between AI performance and power efficiency through a dual-core architecture. The chip uses ARM M52 core for everyday tasks and switches to M55 Cortex for demanding processing, plus an on-board NPU to accelerate Machine Learning and AI algorithms.

Exploration of the BK7259's NPU capabilities enabling edge intelligence—running small language models, facial recognition, voice recognition, and voice commands locally without cloud connectivity. This enables privacy-preserving, low-latency interactions with devices.

Diana details how RiseLink positions itself at the intersection of hardware and conversational AI with voice-ready features including integrated audio ADCs, on-device wake-word detection, noise suppression, and ultra-low Wi-Fi keep-alive. Partnership with Agora enables ultra-low latency and barge-in capabilities for natural, real-time dialogue.

Diana shares her passion project building a voice device for early childhood development—an interactive toy that introduces words through dialogue-driven stories. She envisions a world where intelligent connected devices like e-bikes can hear voice prompts and guide users, eliminating the need for phone navigation.

Overview of RiseLink's three chip categories: BK7259 for edge AI applications running vision and language models locally, BK7258 multimedia SoC for easy audio/video connectivity, and connectivity chips like BK7239 for ultra-low-power Wi-Fi/Bluetooth mesh networks supporting up to 250+ devices.

Discussion on the future of multimodal AI processing visual and audio information simultaneously. Diana explains how on-device memory enables devices to remember previous interactions—critical for companion robots like Fuzozo and adaptive learning toys that remember which words children have learned.

Diana walks through the complete processing pipeline from wake-word detection in Wi-Fi keep-alive mode, local noise filtering and audio processing, to secure cloud transmission when needed for LLM responses, and finally text-to-speech on-device. She explains how intelligent decisions minimize latency and API costs.

Deep dive into BK7259's powerful edge processing—running smaller language models on-device, unlocking smart door locks via facial recognition in under 200 milliseconds (one-fifth of a second), voice commands, and health wearables processing sensor data to detect abnormalities, all without cloud connectivity.

Diana discusses safety considerations for kid-safe devices from both hardware (no sharp edges, no small parts) and software perspectives. She explains her toy's design as an interactive reading engine that fosters dialogue, teaches abstract concepts like emotions, and helps children express themselves—addressing screen time concerns.

Diana shares collaboration with her Yale PhD advisor on early childhood development, adapting their award-winning 10-year training program (proven to improve language ability in 4-6 weeks) into an affordable AI toy. This democratizes access for working parents and lower-income families who can't afford expensive workshops.

Diana encourages developers to reach out directly and start with RiseLink's SDKs and developer kits. The BK7258 dev board comes ready to test with eyes and can be disassembled for custom projects. She highlights GitHub resources and engineering support available to help developers bring their ideas to life.

Diana shares examples of local builders creating intelligent Rubik's cubes with personalities and 3D-printed Halloween spiders powered by RiseLink chips. She discusses plans for hackathons, university events, and supporting the maker community in overcoming hardware development costs and challenges.

Discussion on the importance of dogfooding—Diana's hands-on experience building her AI toy revealed challenges developers face including AI agent design, model selection, token optimization, noisy environment performance, age-based voice recognition, and 3D printing considerations for electronics placement.

Diana envisions the next 1-3 years seeing chips running surprisingly powerful models locally without cloud LLMs, multimodal AI processing visual and audio together becoming prevalent, and memory technology advances enabling chips to store extensive personalization data so devices truly know their users.

Diana and Hermes explore a future where everyday objects have voices and intelligence—intelligent two-wheelers responding via voice, coffee machines announcing when cups are filled, smart fridges warning about spoiled food and inventory, and sending reminders to buy more vegetables.

In response to the wildcard question about what she'd do if not at RiseLink, Diana expresses deep interest in AI education and revolutionizing the toy industry with conversational AI devices that help children learn through dialogue instead of screens—exactly what she's building with her passion project.