Sivaramakrishnan Subramanian

Software Engineer

Carnegie Mellon University•  August 2022 - Present

- Investigating controllable-GAN pipelines with self-supervision for extraction of 3D object-aware representations from tomography images of Cryo-EM structures - Mentoring 2 interns in unsupervised domain adaptation for cell organelle detection in 3D tomogram slices

Software Engineer

AppOrchid Inc•  October 2019 - July 2022

- Designed a MLaaS framework (machine learning-as-a-service) for model in-loop development, training, deployment and serving - Synthesized a custom layout analysis model with Yolo v4 backbone for extracting legalese documents’ structural metadata; built robust (0.73 mAP) and 8x higher throughput model than caffe2 baseline - Prototyped an analysis-by-synthesis pipeline for underline detection(U-Net) and signature detection (spatial pyramid pooling); shipped model had 0.93 precision at 34 fps - Architected a Cascade mask r-cnn model with BiFPN fusion for tabular structure extraction from financial documents; achieved 0.48 AP on ICDAR cTDaR task and reduced latency by 3x after FP quantization