Deep Learning Engineer at Insta360

Researched and implemented vision-language models for video understanding. Fine-tuned and deployed multimodal CLIP-like models such as MobileCLIP on edge devices, achieving ~10x faster inference and ~6% higher accuracy on internal benchmarks. Improved action recognition by ~8% by adapting various VLMs for video understanding. Led an agentic AI-powered video editing workflow to autonomously identify, extract, and assemble highlight-worthy segments from large volumes of raw footage. Maintained a million-video dataset with rigorous quality control pipelines. Created a video style-transfer workflow using Stable Diffusion, LoRA, and ControlNet; developed local object style transfer by integrating SAM and XMEM for mask tracking and local target creation. Optimized UNET with TensorRT, reducing inference time by 10~30%. Implemented traditional vision algorithms and lightweight model optimization (distillation, pruning) to accelerate MobileNetV2 and TSM.

Silicon Validation Engineer (Internship) at Intel Corporation

Ensured protocol integrity (SPI/I2C) for pre-silicon ASIC designs through rigorous validation. Engineered C++ test suites to efficiently evaluate emulator communication on FPGA platforms. Improved technical documentation and cross-team collaboration.

Deep Learning Engineer at Insta360

Led research and deployment of vision-language models for video understanding, including fine-tuning CLIP-like models on edge devices, achieving faster inferences and improved accuracy. Developed agentic AI-powered video editing workflows to autonomously identify and assemble highlight-worthy segments from large footage collections, and maintained a million-video dataset with rigorous quality-control pipelines. Pioneered a video style-transfer workflow using Stable Diffusion variants, integrated SAM and XMEM for local mask tracking, and optimized UNet-based architectures with TensorRT. Also developed an OpenCV-based whiteboard correction algorithm for video conference products and applied model distillation/pruning to create lightweight models with substantial inference-time reductions.

Silicon Validation Engineer (Internship) at Intel Corporation

Ensured protocol integrity (SPI/I2C) for pre-silicon ASIC designs through rigorous validation. Engineered C++ test suites to efficiently evaluate emulator communication on FPGA platforms. Improved technical documentation and cross-team collaboration to streamline validation workflows.

Deep Learning Engineer at Insta360

Researched and implemented vision-language models for video understanding, fine-tuned and deployed multimodal CLIP-like models on edge devices, and led a data-driven video editing workflow. Maintained a million-video dataset with quality-control pipelines; contributed to video style-transfer projects using Stable Diffusion, LoRA, and ControlNet; optimized models (UNET) with TensorRT for faster inference; and developed lightweight vision algorithms for real-time applications.

Silicon Validation Engineer (Internship) at Intel Corporation, Canada

Ensured SPI/I2C protocol integrity for pre-silicon ASIC designs, engineered C++ test suites to validate emulator communication on FPGA platforms, and improved technical documentation and cross-team collaboration.