AI Platform for Single Cell Taxonomy

Cell morphology data sets visualization

In my role as a senior machine learning engineer at Deepcell, I played a pivotal role in developing an advanced AI platform aimed at empowering researchers to analyze and classify cellular tissue composition through the examination of cell morphology. Collaborating closely with data scientists, I fine-tuned deep neural models on the Google Cloud Platform, employing diverse network architectures for the sorting, cleaning, and classification of cell images. To enhance computational efficiency, I leveraged preemptible TPUs and created a hyperparameter tuning tool, ensuring effective scheduling and monitoring of training performance.

In addition to my contributions, I implemented TensorRT and ONNX model formats to facilitate model compression and translation, further optimizing the platform. The development also involved the creation of a model management service, delivering increased value to end-users. To address the challenge of working with vast amounts of legacy embeddings, I designed an index system, making them compatible with the latest deep learning network architecture. Furthermore, I established a robust data pipeline responsible for reading image data from an in-house instrument, applying computer vision augmentation, and pre-processing it into Tensorflow embeddings. The seamless integration of our proprietary instrument with cloud computing resources proved instrumental in assisting biologists and cross-functional teams in their exploration of novel cell cluster types, providing an innovative solution for advancing scientific research.