Summary
Work History
Education
Skills
Websites
Masters thesis
Research
Projects
Timeline
Generic

Kunal Mehta

Austin

Summary

Dynamic Software Engineer with a proven track record at Oracle, skilled in Python and Java. Successfully built predictive models and innovative tools, enhancing performance analysis and reducing bottlenecks. Adept at collaborating in teams and delivering impactful solutions.

Work History

Software Engineer II

Oracle
Austin
07.2023 - Current
  • Extracted and preprocessed data, and built statistical models for performance analysis, thereby identifying the questionable entries.
  • Built a Java-based visualizer tool that extracts data, generates several graphs, and sends email notifications about the cluster's performance.
  • I researched and proved why generative AI could not be used for document-to-communications model migration, and I helped in building a rule-based system.
  • Produced a detailed research study on Oracle Heatwave GenAI, showcasing in-depth analysis and insights.
  • Built several facts and identified issues in fact services.
  • Built two services using Oracle APEX and an interface that can interact with the tables in Oracle APEX.

Software Engineer II Intern

Cisco Meraki
Boulder
01.2023 - 04.2023
  • Built a Random Forest model that would predict the probability of watchdogs getting triggered for different devices
  • Extracted, preprocessed the data, and used the Hugging Face summarizer to summarize the top issues faced by the customers.

Software Engineer Intern

Oracle
05.2022 - 08.2022
  • Analysed log data and identified relationships between different parameters, thereby finding large blobs, missing indexes and large number of small network connections as the three primary causes of bottlenecks
  • Built and trained Decision Tree and Random Forest models with 89% accuracy to predict elapsed time and identify the causes of bottleneck, respectively
  • Designed, built and deployed a system on OCI that preprocesses log data from the Elasticsearch, allows engineers to label the unconfident samples predicted by the Random Forest model in Flask server and thereby storing in MySQL database
  • On labelling 1000 samples, the model is be retrained on this data and replaced if it has a higher accuracy

Associate Web Developer

Media.net
01.2021 - 08.2021
  • Implemented RESTful API leveraging the Lumen framework that serves requests and filters keywords, sets TTL, and generates a response to the user & caches on Redis before storing it on the MySQL database
  • Collaboratively built a Request analyzer using Elasticsearch(ELK stack), that identifies the bot requests
  • Analyzed log data, and designed and engineered microservices that detect the presence of bots in the incoming requests, thereby storing them in the Elasticsearch
  • The request analyzer has reduced the bot traffic by 40%

Student Developer

Orcasound
06.2020 - 09.2020
  • Company Overview: Google Summer of Code 2020
  • Built an Active Learning tool that identifies low confidence Orca calls using the uncertainty sampling technique
  • Having saved approximately 800 hours of labeling time, the tool is used by more than 50 citizens and 12 bioacousticians
  • Designed a microservice architecture consisting of CNN model that queries and identifies uncertain calls from the S3 bucket
  • The location of uncertain samples is stored in PostgreSQL database and passed to the user for labeling
  • Proved that retraining the CNN model only on uncertain samples achieves similar accuracy as if it was trained on the entire data set
  • Containerized the application using Docker and hosted on AWS Lightsail, with frontend in React and Typescript and backend in PostgreSQL, and an API in Flask that serves as an interface between machine learning model and web app

Research Intern

IIT Bombay
10.2019 - 05.2020
  • Developed an intuitive language called Spec DFA by adopting syntax similar to Golang and created a transpiler that transpiles input analysis in Spec DFA to its Python equivalent
  • Introduced macros, short specification, long specification, and new data types in the input language
  • The transpiler has allowed more than 300 users to generate python code by learning the intuitive spec DFA language

Education

Masters of Science - Computer Science

University of Colorado Boulder
Colorado
05.2023

Bachelor of Engineering - Computer Engineering

K.J Somaiya Institute of Engineering and I.T, University of Mumbai
10.2020

Skills

  • Python
  • Java
  • C
  • MySQL
  • PostgreSQL
  • HTML
  • CSS
  • Typescript
  • JavaScript
  • PHP
  • React
  • NodeJs
  • Flask
  • Docker
  • REST
  • AWS
  • TensorFlow
  • Keras
  • OpenCV
  • NLTK

Masters thesis

Designed and implemented a distributed object detection framework comprising frame extraction, concurrent object detection (via multiple Dockerized containers in a Kubernetes cluster), and frame combination.

Evaluated system performance by testing various configurations—adjusting the number of object detection servers and video quality—resulting in significant processing time reductions.

Analyzed scalability and optimization strategies for real-time object detection applications, providing insights to enhance accuracy and efficiency for fields like security, autonomous driving, and robotics.

Research

Orca call detection using CNN and Spectrograms, 10/19, 03/20, Designed a CNN model and showed its effectiveness when combined with template matching, allowing us to reach an accuracy of 92% even after being trained on a limited dataset (https://ssrn.com/abstract=3572303)., Compared the accuracy with the following machine learning models: HMM, GMM, RNN, ResNet-512 and found CNNs to be the most effective.

Projects

Mastek's Deep Blue Plastic waste brand detection, Keras, Tensorflow, scikit-learn, Python, Git, Spearheaded a team of two, using Agile methodology, trained Faster R-CNN model that identifies 45 different brands of plastic with 86% accuracy., The model can differentiate between plastic and cardboard wrappers of the same brand having similar images., The Android application has been used by around 50 students and has significantly reduced plastic waste near University. E-commerce, PHP, CakePHP, MySQL, HTML, CSS, Git, Devised an eCommerce website leveraging CakePHP, an MVC framework, with search and order placement functionality., Developed standalone backend and frontend connected via an API interface and wrote automated tests using Selenium., Wrote unit tests and integration tests, GitHub actions, and created a CI-CD pipeline for automatic deployment.

Timeline

Software Engineer II

Oracle
07.2023 - Current

Software Engineer II Intern

Cisco Meraki
01.2023 - 04.2023

Software Engineer Intern

Oracle
05.2022 - 08.2022

Associate Web Developer

Media.net
01.2021 - 08.2021

Student Developer

Orcasound
06.2020 - 09.2020

Research Intern

IIT Bombay
10.2019 - 05.2020

Masters of Science - Computer Science

University of Colorado Boulder

Bachelor of Engineering - Computer Engineering

K.J Somaiya Institute of Engineering and I.T, University of Mumbai
Kunal Mehta