Camera AI Engineer (IoT)

## Company: Qualcomm China ## Job Area: Engineering Group, Engineering Group > Software Engineering General Summary: Job Overview Artificial Intelligence is changing the world for the benefit of human beings and societies. QUALCOMM, as the world’s leading mobile computing platform provider, is committed to enable the wide deployment of intelligent solutions on all possible devices – like smart phones, laptops, autonomous vehicles, robotics and IoT devices. Qualcomm is creating building blocks for the intelligent edge. We are seeking a passionate and hands-on Machine Learning Engineer to join our AI Software team in China. You will support Model Onboarding requests for Qualcomm’s China customers, helping take customer and third-party models from initial request to production-ready deployment on Qualcomm Snapdragon platforms. Workloads span the full model spectrum – from large language models (LLMs) and generative AI models to latency-critical vision, audio, and multimodal networks. The role is grounded in Qualcomm’s AI technology stack: Snapdragon SoCs, the Hexagon NPU (HTP), Adreno GPU, and Qualcomm’s AI software and SDKs including QAIRT (Qualcomm AI Runtime SDK). You will use these tools to analyze, quantize, convert, and deploy models, and to debug and fine-tune their accuracy and performance on simulation and on real hardware. Roles & Responsibilities • Support Model Onboarding requests from Qualcomm customers in China, helping drive each request from intake through production deployment across LLMs, generative AI, vision, audio, and multimodal models. • Analyze and adapt customer models: review architectures (CNNs, transformers, LLMs, diffusion, and multimodal networks), identify unsupported operators or structures, and adapt models for efficient execution on Qualcomm hardware. • Quantize, convert, and compile models from common frameworks (PyTorch, TensorFlow, ONNX) into Qualcomm runtime formats using QAIRT and related Qualcomm AI tools, applying PTQ/QAT and LLM/GenAI techniques (weight-only, activation, and KV-cache quantization, mixed-precision; Lora, Batch, etc) to meet targets. • Validate model inference via simulation (x86 reference) and on device, checking numerical correctness and functional behavior on the Hexagon NPU (HTP). • Debug and fine-tune accuracy and performance: analyze quantization error and layer-wise mismatches, profile latency, memory, and power (including LLM token rates and KV-cache footprint), and apply hardware-aware tuning to meet customer KPIs. • Collaborate with Qualcomm model, tool, framework, and SDK teams to align onboarding solutions with SoC and HTP capabilities, and to escalate gaps in operator or feature support. • Participate in the Software Development Life Cycle: requirement analysis, design reviews, development, testing, integration, debugging, and performance tuning. Required Qualifications • 3+ years of hands-on experience in machine learning model deployment, optimization, or embedded AI development on IoT, mobile, or real-time embedded platforms. • Strong proficiency in Python and C/C++, including performance-critical, system-level programming and ML tooling workflows. • Solid understanding of deep learning model architectures – CNNs, transformers, LLMs, diffusion, and multimodal models – and common frameworks such as PyTorch, TensorFlow, and ONNX. • Knowledge of model optimization techniques – quantization (PTQ/QAT), pruning, graph optimization, and hardware-aware model design – with strong analytical and debugging skills for accuracy and performance issues on real hardware. Preferred Qualifications • Hands-on experience with Qualcomm AI SDKs – QAIRT (Qualcomm AI Runtime SDK) or Hexagon SDK – and with deploying models on Snapdragon using the Hexagon NPU (HTP) or Adreno GPU. • Experience optimizing or deploying LLMs and generative AI models (e.g. LLaMA, Qwen, Mistral, Stable Diffusion, or similar VLM/LMM families), including weight-only and KV-cache...