Avinash Anand

Background

About Avinash Anand

As part of Qualcomm’s XR (Extended Reality) team, I work at the intersection of AI, edge computing, and immersive technologies to enable intelligent XR experiences on next-generation headsets and wearable devices. My responsibilities include: Model Onboarding for XR Use Cases • Collaborate with research and product teams to identify suitable deep learning models (e.g, vision, perception, tracking, hand/eye gesture recognition) for XR workloads. • Adapt and convert models (e.g, from PyTorch/ONNX/TensorFlow) into formats compatible with Qualcomm’s inference stack. • Perform quantization (e.g, INT8) and evaluate accuracy trade-offs for real-time XR deployment. • Ensure robustness of models under XR-specific constraints like low latency, limited compute, and dynamic environments.Optimisation for Hexagon DSP (HTP) Backend • Profile and optimize DNN models to run efficiently on Qualcomm’s Hexagon Tensor Processor (HTP), targeting low-power real-time inference. • Leverage Hexagon SDK and Qualcomm AI tools (e.g, QNN) to write custom DSP kernels and utilize SIMD/HVX/HMX instructions. • Analyze bottlenecks and apply memory, tiling, and vectorization strategies to maximize performance and minimize latency. • Work closely with firmware and systems teams to align AI workloads with the broader XR pipeline and power budgets.Performance Tuning and Validation • Benchmark models across multiple XR hardware platforms. • Tune performance KPIs including FPS, power, and thermal impact. • Validate model behavior across different runtime conditions (e.g, VIO drift, gesture noise, lighting variability)