Atul RAghunathan

NLP Researcher. Cloud Architect. Leader.


15-316 Software Security and Privacy

11-411 Natural Language Processing 

15-251 Great Theoretical Ideas of Computer Science 

15-210 Parallel and Sequential Data Structures and Algorithms

15-213 Computer Systems

36-218 Probability Theory

15-150 Functional Programming

15-122 Imperative Computation

10-315 Machine Learning

36-218 Probability Theory

36-401 Modern Regression

15-281 AI Representation and Problem Solving 

I am currently an undergraduate at Carnegie Mellon University with a passion to change the world for the better with technology. 

My expected graduation date is in December 2021.




– Certified by AWS  in August 2019
– Currently working towards Solutions Architect Certification
– Lockheed Cloud Deployment and Chaos Engineering

All data annotation/processing for machine learning work for:
Patent Classification and Interpretation

Developed AMI for cloud-based Jupyter Notebook Processing. Used by several ML and NLP research groups at CMU. 

All Natural Language Processing work for:
Patent Classification and Interpretation

I wrote Windows applications in Windows Presentation Foundation during my time at Lockheed Martin. For security reasons, I cannot discuss the content of the applications.

Current Work

Patent Classification and Interpretation

I am working with Professor Dean Alderucci to develop a Machine Learning and Natural Language Processing framework in Python to categorize and extract technical information from the USPTO patent database. I am on track to publish my work within the year. I used spaCy as my dependency parsing engine to identify Beauregard Claims with a multi-tier binary classifier with 93%+ accuracy. I then utilized dependency parsing to break the claim into several subsections based on the four key parts of a Beauregard Claim, and found the purpose of the software described by the claim. I am currently working with Google's BigQuery to develop a machine learning model based off of datasets generated by my Binary Classifier, to more expeditiously classify large sets of data.

Please find my preliminary report here: Report

More information coming soon! I’m working towards a Publication.

Feel free to use the contact form down below to email me with any further questions.



Software Development Engineering Intern in Amazon Search

I developed transfer learning strategies to improve the performance of Amazon Search in both emerging and established marketplaces.

I utilized data from several sources to build language models to supplement current ranking and classification strategies used in search. Internal testing indicated significant improvements in 6 marketplaces.

I created cloud-based flexible compute infrastructure to create language models from massive training corpora. I utilized Spark for EMR operations and AWS Batch for sequential operations, while allowing for easy integration with Amazon's dependency management system. In addition to helping tune hyperparmeters for transfer learning, this infrastructure was adopted by the team for rapid language model development and experimentation.

Software Engineering Intern in Experimental Technologies

I led a team with Agile to develop a highly-available cloud infrastructure as code. This included a multi-AZ Autoscaling group and Database with a fully-managed elastic load balancer. This deployment was designed to withstand DDoS, Resource, State, and Network attacks. To verify these automated recovery policies, I implemented chaos engineering with Gremlin.

Additionally I worked on a project to integrate a Docker-in-Docker test environment in AWS GovCloud with Lockheed Martin's secure Docker Image registry. The system was later used to train employees on Docker.

I also built a parallel framework for physically locating thousands of internet connected devices in a corporate environment to improve security. I packaged this solution with a command line interface and all necessary libraries into a container image for future use and portability.

Feedback Control of Advanced Lighting Systems

Under the guidance of Professor Sandipan Mishra at Rensselaer Polytechnic University, I wrote software in Python, Matlab, and C++ for the Intelligent Automation and Control (ISAaC) lab to dynamically adjust internal lighting conditions based on a matrix of color sensor readings in a space.

With a cost function that weighs color equality, electrical cost savings, and light temperature aesthetic appeal, a multivariate optimization was applied with gradient descent. Since the lighting fixtures were RGBAW and the color sensors were RGB, the software controller utilizes the two degrees of freedom to further minimize the cost function.

Redirected Walking

I worked with Professor Brett Jackson at Carelton College to further develop a Virtual Reality (VR) technology known as redirected walking. Redirected walking can "redirect" a user using a VR headset in a limited physical space to give the interpretation of a larger virtual space.

I contributed to this research by leading a group of interns to investigate the limits of human inertial sensory input, and using this data to develop ideal applications for redirected walking technology in Unity. One demo that we developed reduced required physical area by 400%, demonstrating the effectiveness of this technology.

To access experimental non-VR demo, download contents of this folder: Demo and run “Redirected Walking Build.exe”. Control simulated user motion with arrow keys. May not work as intended on all systems. 

Six Degree Positioning Based on Laser Plane Projection and Inertial Navigation

This project provides 3D spatial orientation in an unknown space through the establishment of thermal patterns (beacons) on objects within the environment. Utilizing the skew, rotation, and size of the pattern, this method can derive six degree of freedom data (x, y, z, pitch, yaw, roll). Positional XYZ data was consistently accurate within 5 centimeters and degree data was accurate within 2 degrees. I used two co-processors in this project: an Atmel based computer to control laser activation and gimbal movements, and an Linux based (Raspberry Pi) to read IR camera information and calculate Six Degree Positioning.

1st Place Synopsys Santa Clara Regional Science Fair

Yale Science and Engineering Award

4th Place California State Science Fair

Honored by Mayor of Cupertino

Abstract: CSSF Abstract

Project Journal: Journal

Computer Vision/Signal Processing: Algorithms Overview

Recovering and Recycling Excess Thermal Energy to Improve the Net Efficiency of Common Lighting Sources

Even the most efficient consumer light form, the LED, converts between 5-40% of its energy input into heat. Everyday household LED bulbs have large radial heatsinks to remove and dissipate this heat. This project aims to utilize this wasted heat by collecting it, converting it into electrical energy, and recycling that energy to partially power the LED bulb. With a combination of inexpensive thermoelectric generators, capacitors, and transistors, such bulbs can save between 129 and 322 kilowatts of power over their lifetime when instituted in the average american home.

1st Place Synopsys Santa Clara Regional Science Fair 

Silver Medalist I-SWEEEP International Science Fair

IEEE Electro-Technology Award

2nd Place California State Science Fair

Abstract: CSSF Abstract

Circuit Diagrams: Circuit Diagrams

Project Journal: Journal


Deewane Acappella

In addition to performing, I currently serve as the tour manager and web developer for my Acappella group at Carnegie Mellon University. We compete in local and regional competitions as a part of a national Acappella circuit. Check us out on Spotify!

Cupertino High School Robotics

As the Project Manager/Captain of my 75+ person FRC team (Team 2473), I led the delivery and integration of robot software and hardware. With Agile management principles, and Gantt based scheduling, I was able to keep 10 separate sub-teams on track throughout development and training seasons.

I led a LiDAR research sub-team (FRC 2473) to investigate the use of real-time LiDAR based object and feature detection. I developed algorithms in Python for corner detection and deducing robot position on field. By inspecting the r-values for linear regressions across the FOV of the 2D LiDAR system, our team was able to find the probability of corner at a given azimuth.

I led a Six Degree Positioning research sub-team (FTC 4950) to find robot location and angle (over short time periods) based on an Inertial Measurement Unit. This project was written in Java and compiled to an Android Application and used for Autonomous robot navigation. Our team was awarded the Control (Software) Award during competition.


Let me know what you think! I'm currently seeking software engineering intern and networking opportunities.

© 2019 Atul Raghunathan