• Spearheading the development of cutting-edge AI agents at Synth AI, automating optimization and enhancing efficiency.• Researching AI solutions to accelerate the production of AI agents, optimizing performance and scalability.• Driving advancements in AI technology, positioning Synth AI as a leader in the industry.

Statistical Analysis for Machine Learning models with the Fire TV Science team to train models for Time of Day aware recommendations taking genres of content into consideration. Data was based on watch data of when customers watched titles. Concurrently, worked on Partner Integration team. Work on FireTV devices (Android devices) to transmit watch data from Providers (HBO, Netflix, etc.) to AWS servers for the Continue Watching and related features.

I am a Data Scientist and Software Engineer at Amazon FireTV. I use machine learning for entity matching (also known as pairwise matching) on live production systems. The work I do matches different media items together (movies, TV episodes, etc.) from different data providers (HBO, Netflix, Amazon Prime, etc.) and clusters them for use as a data catalog, used by search. The model works to match and cluster tens of thousands of media items prioritizing precision.

Took class and completed coursework for XCS224 - Natural Language Processing with Deep Learning. Learned about topics including transformers, pre-training, NLP, word2vec, etc.

Worked on machine learning techniques for computer vision and reinforcement learning for robotic grasping.

Worked in the Caltech CAST (Center for Autonomous Systems & Technologies) research lab. Worked on a project where we use imitation learning to speed up finding initial solutions for SATO (Sequential Assignment & Trajectory Optimization) using a tree based search RRT (Rapidly-exploring Random Tree). Specifically, we use imitation learning to develop an estimated cost function to bias which branches of the tree to explore. The intuition behind this is that we want to speed up searching for optimal paths for each drone (given multiple drones) before we need to ensure drone collisions do not occur by passing each drone path into SATO. Gained experience with imitation learning, pytorch, and planning with dynamics (double integrator) for multiple robots.

Worked in the University of California at Berkeley AUTOLab with artificial intelligence and robotics. Specifically I worked on two projects. The first is Semantic X-Ray, which is a collaboration with my old team at Stanford University in the Stanford Vision and Learning Lab (SVL). For this project, we are using both semantic and geometric information in our environment to uncover a target object. Specifically our environment is a house, where we navigate through using a mobile Fetch robot using these sources of information. Eventually, we reach a shelf environment, where we need to grasp an occluded object. Throughout the exploration of the environment and reaching for the object on the shelf, we combine both semantic and geometric information in unison. The second is LAX-Ray (LAteral X-Ray), where we extend the X-Ray algorithm to take depth into account. X-Ray originally predicts the distribution of where a partially or fully occluded object is given the other objects in the image. However, these objects are dropped on a flat surface, so the size of the objects don't really change. In this work, we investigate using a shelf environment, where there's significant depth. So we account for this. Additionally, we use a policy that uncovers the target object using solely robot arm pushes, not removing any objects from the environment. The contributions include that we are doing mechanical search in this new environment using pushes (not removing objects) using a novel policy and extended the previous: X-Ray: Mechanical Search for an Occluded Object by Minimizing Support of Learned Occupancy Distributions. . Note: Both projects extend this paper.

Author on the paper: Visuomotor Mechanical Search: Learning to Retrieve Target Objects in ClutterPublished in IROS (International Conference on Intelligent Robots and Systems) Conference 2020Worked in the Stanford Vision and Learning Lab with artificial intelligence and robotics. Specifically I worked on the project Learned Mechanical Search, the goal of which is an intelligent multi-step retrieval of a partially occluded object in a physical bin using a robotic arm. This is accomplished through reinforcement learning using a Sawyer robot arm with an Xbox Kinect visual sensor. I gained experience with reinforcement learning, ROS, robotic systems, Python, PyBullet, and computer vision.

Worked in the Caltech Vision Lab on a project where we worked on using machine learning based image classification to detect dementia using fundus retinal images. I used convolutional neural networks and machine learning along with image filtering techniques. Gained experience with image classification, deep learning, convolutional neural networks, python and PyTorch.

Teaching Assistant for the classes CS/EE 156a: Learning Systems and CS/EE 155: Machine Learning and Data Mining. I grade problem sets, hold office hours and recitation sessions, and manage online discussion forums for helping students understand concepts in machine learning.

One versus One Gameplay for the Distributed Autonomy / Remote Control Research group. Used the Deep Q Learning Reinforcement Learning algorithm train an autonomous vehicle to beat an opponent in arealistic competitive game, with a large number of parameters. Also investigated Genetic and Evolutionary Reinforcement Learning Algorithms for this problem. Worked closely with employees, optimized code for allowing many parameters while keeping the complexity ofthe program small due to the high complexity of the problem. Gained experience with Reinforcement Learning, Deep Q Learning, Q Learning, Python, Tensorflow, Numba.

Microcontroller Attached as Redundant Check for an Unstable Spacecraft (MARCUS) Subsystem. Programmed small CSCI (Computer Software Configuration Item) in C++ for an Atmel Microcontroller to communicate with other components of a spacecraft. Worked closely with employees, constructed framework and performed regression testing. Gained experience with C++, C++ Templates, Atmel Microcontrollers, Atmel API’s, Python, communication protocols,and hardware interrupts.