Company Overview
An IC design start-up founded by senior leaders from Huawei and IBM, specializing in advanced wireless ICs including UWB, BLE, NB-IoT, and WiFi6, with a strong track record in wireless IC development and mass production.
Job Summary
Lead the optimization and deployment of AI audio and video models for resource-constrained IoT devices, driving hardware-aware acceleration and end-to-end performance improvements for next-generation AIoT chips.
Responsibilities
- Lead end-to-end optimization of audio (wake-word detection, ASR, acoustic event detection) and video (object detection, recognition, segmentation) AI models for resource-constrained IoT devices to improve efficiency and accuracy
- Apply quantization techniques (INT8/INT4/mixed-precision, PTQ & QAT), pruning (structured/unstructured), knowledge distillation, and other compression methods to reduce model size, latency, and power consumption while maintaining accuracy
- Optimize models for heterogeneous hardware platforms (NPU, DSP, MCU) using frameworks such as RKNN, Hexagon, Ethos, and TFLite Micro to enhance hardware performance
- Develop custom operators, implement operator fusion, and perform memory planning and cache optimization to maximize hardware utilization
- Conduct hardware-aware quantization and Neural Architecture Search (NAS) to tailor models for specific AIoT chip architectures
- Build automated pipelines for model optimization and deployment to streamline production workflows
- Optimize multimodal (audio + video) models for real-world IoT applications, ensuring best-in-class latency, throughput, power consumption, and memory footprint on production devices
- Research and implement state-of-the-art techniques in model compression, quantization, and dynamic inference to maintain technological leadership
- Collaborate closely with hardware and system teams to enable software-hardware co-design for next-generation AIoT chips
Required competencies and certifications
- Master's degree or higher in Computer Science, Artificial Intelligence, Electronic Engineering, or related fields
- Minimum 3 years of hands-on experience in on-device AI model optimization with successful production deployments
- Proven expertise in at least 2-3 of the following areas: quantization (PTQ, QAT, mixed precision), pruning (structured & unstructured), knowledge distillation (single/multi-teacher, feature distillation), model compression (sparsity, low-rank decomposition), dynamic inference (Early Exit, BranchyNet, conditional computation), hardware acceleration (NPU/DSP optimization, operator fusion)
- Proficiency in AI frameworks such as TFLite / TFLite Micro, ONNX Runtime, TensorRT, RKNN, SNPE, or similar
- Strong programming skills in C/C++ and Python
Preferred competencies and qualifications
- Experience optimizing audio models (KWS, ASR) or vision models (YOLO, MobileNet, EfficientNet, etc.)
