Sriram Vasudevan Email & Phone Number

I am Sriram Vasudevan, a scientist in Advanced development group at Hughes network systems. My primary job description is to use machine learning to improve satellite network systems. Over the last few years, I have garnered experience creating multiple ML models from inception to deployment. I have worked on all stages of the model pipeline such as scaled data collection, preprocessing, feature selection, model architecture design, hyper-parameter tuning, model optimization for deployment, and automated model performance monitoring and retraining after deployment. Over the past year, I built recurrent neural network based models to predict video streaming experience of customers by inferring objective metrics like video resolution and video buffer health from raw traffic data. The model could successfully track the status of customer's video player, and quantify their subjective video streaming experience. This model also played a pivotal role in improving Hughes’s bandwidth allocation algorithm and reducing customer churn. Prior to that, I created ML models to perform network traffic categorization based on QoS classes like interactive, streaming, bulk upload, download and control. This model led to an improved network traffic queueing system that would account for QoS requirements of the application while enqueuing packets. I also played a key role in the optimization and deployment of these models into an embedded system. I am currently working on constantly monitoring and retraining these models on new data. Aside from ML projects at my workplace, I like to indulge myself with small toy ML projects in my free time. Last year, I built a dashboard camera-based vehicle speed detection system and a frontal face based age and gender prediction models. Links to these models and the code are available on my Github profile. Previously, I finished my Masters at University of Maryland, College park, specializing in Machine learning and Computer vision. I was a Graduate research assistant at Prof. Rama Chellappa's lab, where I built UNet-like convolutional neural networks for material classification using multi-spectral satellite images. One of the primary challenges in that project was the lack of adequate labeled data, which I solved by building a UI tool using OpenCV. Prior to that, I interned at US Army research laboratory, where I finetuned a Segnet based convolutional neural network to perform semantic segmentation on Army datasets. I believe my skills and prior experience makes me a good fit for companies operating in this domain.