Zirui Song

zrsong@amazon.com

I am a Machine Learning Engineer at AWS, working on the OpenSearch project. My work focuses on neural sparse retrieval and approximate nearest neighbor search, where I contributed to the integration of the SEISMIC algorithm into OpenSearch — enabling efficient semantic search at billion-document scale with sub-3ms latency. The feature shipped as Neural Sparse ANN in OpenSearch 3.3.0, achieving 17–35x latency reduction and 18x throughput improvement. I also contribute to Neural Sparse Embedding model fine-tuning for improved retrieval quality.

I obtained my Master of Engineering in ECE from Cornell University in 2024, where I worked with Prof. Qing Zhao on federated learning with adaptive edge device selection. Before that, I received my Bachelor of Engineering in EIE (First-Class Honors) from The Hong Kong Polytechnic University in 2023, where I was supervised by Prof. David Navarro-Alarcon at the ROMI Lab on robotic ultrasound imaging, and by Prof. Kenneth Lam for my final year project on archived film restoration. I also spent a summer as a research assistant at Sharc Lab, Georgia Tech, supervised by Prof. Callie Hao, working on federated learning optimization.

My interests span Information Retrieval, ML Systems, Cache-aware Optimization, and Applied Deep Learning.

news

Mar 17, 2026	Gave a speaking session of Scaling Semantic Search to Billions with Neural Sparse ANN in OpenSearchCon China 2026.
Mar 15, 2025	Neural Sparse ANN shipped in OpenSearch 3.3.0, enabling sub-3ms sparse vector retrieval at billion-document scale.
Jan 06, 2025	Joined AWS OpenSearch as a Machine Learning Engineer, working on neural sparse retrieval and search systems.
Jun 24, 2024	Our paper “A Dynamical System Approach to Robotic Ultrasound Imaging” was published at IEEE RCAR 2024.

selected publications

IEEE RCAR
A Dynamical System Approach to Robotic Ultrasound Imaging: Towards Intrinsically Stable Robotic Sonography

Wanli Liuchen, Anqing Duan, Zirui Song, Maria Victorova, and David Navarro-Alarcon

In 2024 IEEE International Conference on Real-time Computing and Robotics (RCAR), 2024

Abs Bib

We present a framework combining ResNet deep learning with time-invariant dynamical systems to enhance robotic ultrasound-guided scanning. Our approach enables robots to fulfill multiple objectives concurrently – sustaining uniform contact pressure and achieving exact positioning – while preemptively adjusting dynamics to maintain stable patient interaction. Experimental validation on a UR5 robotic arm demonstrates stable scanning with force tracking RMSE as low as 2.54N, and the deep learning model achieves 80% accuracy in organ detection.
@inproceedings{liuchen2024dynamical, title = {A Dynamical System Approach to Robotic Ultrasound Imaging: Towards Intrinsically Stable Robotic Sonography}, author = {Liuchen, Wanli and Duan, Anqing and Song, Zirui and Victorova, Maria and Navarro-Alarcon, David}, booktitle = {2024 IEEE International Conference on Real-time Computing and Robotics (RCAR)}, pages = {647--652}, year = {2024}, publisher = {IEEE}, }