Rohit Bandaru
I am a software development engineer at Amazon in the Halo Tone Science team. I work on implementing machine learning to provide tone analysis on user speech. Prior to Amazon, I was at Cornell University where I focused on machine learning and computer vision research. I earned a B.S. and M.Eng in computer science with a minor in electrical and computer engineering.
Visit my blog to read about some interesting topics in AI and computer vision.
Research
At Cornell, I was able to be involved in a variety of research projects, both through PhD level classes and research groups.
Dynamically Adding and Removing Neurons from Neural Networks
The goal of this project was to develop more general methods to learn efficient and effective neural network architectures for certain applications. It is difficult to find effective and efficient architectures for applications. This often leads to networks being larger and more inefficient than they need to be. We are developing an algorithm that can apply iterative pruning while also adding nodes to allow for a neural network to learn a better size and structure.
[Paper] [Code]
Domain Adaptation for FashionNet
Worked with professors Bharath Hariharan and Kavita Bala to create a new dataset of different types of fashion images, and use various domain adaptation techniques to improve performance of the FashionNet model.
Extending Graph Convolutional Networks to Edge Attributed Networks
Developed new architectures for graph convolutional networks (GCNs) to leverage node and edge based features.
[Paper] [Code]
Accurate Kernel Interpolation with Compactly Supported Kernels
This project aims to use compactly support kernels to make scalable KISS-GP Gaussian process framework more accuracte and efficient.
[Paper]
Pancreatic Tumor Classification
This project sought to explore the different deep learning architectures that can be used for the classifcation of pancreatic tumors. Early detection and classification of these tumors is critical for treatment. We developed a model that achieved 69.8% classification accuracy on our dataset.
[Paper]
Seizure Detection
This project employs machine learning for seizure detection using electroencephalography (EEG) data. Detection is an important problem in treating epilepsy. We used time series data, and wavelet transform coefficients with dimensionality reduction on support vector machine and random forest machine learning models. We were able to achieve over 80% area under ROC curve performance.
[Paper] [Code]
Organizations
I have had the pleasure with great organizations including technology companies and project teams.
Amazon.com
Software Development Engineer
I currently work in the Halo Tone Science team. Halo is a health and wellness focused wrist band with an accompanying mobile application. One of its features is tone, which identifies the user's speech and provides tone analysis to help them improve their health. I work on the machine learning workflow to enable this tone analysis. Learn more...
Amazon.com
Software Development Engineer Intern
I worked at Amazon in Seattle to develop a Spring MVC web application for self service configuration of customer service surveys, reducing 1-2 days/week of SDE effort. I was able to learn and contribute a lot with my team. I went beyond my initial project scope by determining business need for additional features that I then implemented.
Cornell Computer Science Teaching Assistant
CS 4670 Computer Vision, CS 4780 Machine Learning, CS 4320 Database Systems
I enjoy helping others learn and share my passion. I am currently working as a graduate teaching assistant for Computer Vision. My duties include helping students with the material and homework by holding office hours, grading class assignments, developing programming assignments, and answering additional student questions on the online forum.
Cornell iGEM
Synthetic Biology Engineering Project Team
I have worked heavily with the Wet Lab, Business, and Product Development subteams at Cornell iGEM. My main focus is develop biological projects into marketable products through entrepreneurship and engineering. Learn more...
Huna Makia
I worked with this enterprise intelligence start up to develop an iOS application to complement their web service by letting job recruiters intuitively search the company's database for potential candidates and effortlessly contact them. I gained significant experience in UX design and mobile development. Learn more...
Cornell Autonomous Bicycle Team
I led the computer vision localization project for the autonomous system to understand its location and surroundings using machine learning and odometry. I worked to setup an effective development system using a Nvidia Jetson TX1, Zed Stereo Camera/SDK, ROS, Python, and a Linux environment. Learn more...
Amazon Halo Tone Science
Cornell iGEM Team
Cornell iGEM is an engineering project team that aims to solve a real world problem using synthetic biology and engineering. We compete in the international competition in Boston every October.
For the 2016 Project, Legendairy, I worked primarily with the wet lab and business subteams. I worked to clone bacteriocin genes into E. Coli bacteria in order to develop a non-antibiotic treatment for bovine mastitis, which is a disease in cows. Although it is not my main field, I enjoy working in bioengineering/synthetic biology and am excited for its future potential. I developed the project entrepreneurally and secured sponsorships on the business subteam. I wrote a business plan and met with people on campus to improve the idea. I also helped with the web development of the wiki.
For the current 2017 project, I trained the new members with wet lab work and continued with the entrepreneurial and sponsorship aspects of the business team. I am also working with the product development subteam to engineer hardware and software for the project.
Huna Makia Internship
I developed EngageApp, an application that allows a user to quickly search the database for a candidate to contact, and leave a voicemail. This application was meant to provide a quick way for job recruiters to search and contact target candidatates.
I began with UX/UI design. I studied the research of BJ Fogg and Nir Eyal to create an optimal user experience that will retain users. I wanted to design an intuitive app that keeps users coming back. I used Sketch to create mockups, and Invision to create prototypes. I used the UserTesting platform to test my prototype with real users. I then used this feedback to improve my designs.
After going through multiple design cycles, I began developing the application for iOS with Swift. Using libraries such as SwiftyJSON, and Alamofire, I was able to write clean and extensible networking code. I also worked with many different development frameworks including AudioKit to record voicemails and MessageUI to send emails within the app.
Cornell Autonomous Bicycle Team
Cornell Autonomous Bicycle Team is an engineering project team composed of a variety of different engineering majors, with the goal of developing an efficient and intelligent self driving bicycle.
I led the computer vision localization project for the autonomous system to understand its location and surroundings using machine learning and odometry. I worked to setup the hardware and software for computer vision development.
I led development of the web application component of the system. This was meant to provide an interface to create a bike route with Google Maps and send the navigation data to the bicycle hardware. It also receives physical debugging data such as speed, position, and lean angle, and stores it in a PostgreSQL database.
The main functions of the web application are navigation and data storage and analysis. The web server sends navigation data to the bicycle using the Google Maps API. It receives data from the hardware and stores it in a PostgreSQL database. There is also an interface to download and interact with the data. This data is physical state data from the bicycle (velocity, lean angle, etc.) and is used to optimize performance.