Master of Science in Computer Science and Engineering from University of Southern California (Aug. 2022 - May. 2024)

Bachelor of Technology in Computer Science and Engineering from MIT World Peace University
(Jul. 2018 - Jun. 2022)

Led USC’s REDCap & KeckCap pages development, web-tools for research databases & surveys, using HTML, CSS, Bootstrap, JavaScript.

Revamped and tested the patient response UI using ReactJS and Next.js for Keck’s appointment scheduling & waitlisting app, enhancing appearance and responsiveness. Used npm for efficient management of project dependencies.

Developed and maintained database schemas, queries, and stored procedures using SQL Server.

Deprecated outdated functionalities & updated Apex classes for Salesforce and Mulesoft integration with Cerner of a Health CRM application.

Analyzed MuleSoft project flows, identified Salesforce object references, and crafted detailed documentation to simplify team understanding.

Achieved a 95% reduction in time required for structured data creation for software testing by automating it using Python scripting.

Executed ETL processes by converting SailPoint employee Excel data through SQL transformations for warehouse-ready integration.

Proactively conducted annotation and data analysis of ServiceNow tickets to identify root causes of increased ticket volumes.

Identified incident patterns in the ticket volumes using data visualization in Excel and Tableau to find key trends for reducing ticket volumes.

Researched and developed a machine learning pipeline using Python, PyTorch, TensorFlow, and Keras to extract and analyze context, while modeling intricate long-term dependencies within English text.

Applied statistical techniques using Google n-grams for context extraction and language modeling.

Used BERT for Next Sentence Prediction and Masked Language Modeling to improve contextual understanding and text generation.

Fine-tuned BERT, Long Short-Term Memory, & Recurrent Neural Networks for Named Entity Recognition & Part-of-Speech tagging.

The Society of Women Engineers is a non-profit dedicated to enhancing quality of life and promoting diversity.

Contributed to the technical department by assisting with the web development of the SWE-MITWPU website.

SWE-MITWPU aims for the comprehensive professional development of its members.

It focuses on preparing students for improved professional prospects, career advancement, and networking opportunities.

Participated in talks and workshops organized by SWE-MITWPU.

Spearheaded the organization, recruitment, and execution of the TEDxMITWPU 2022 event.

Curated speakers, enriching the event with diverse and engaging talks.

Oversaw the operational aspects, guaranteeing the smooth functioning of the event.

Contributed to organizing TEDxMITWPU 2019, participating in marketing, content writing, photography, and operations.

Developed an explainable model for maritime object detection and classification using AlexNet, ResNet101, VGG16, and DenseNet161.

Utilized LIME, GradCAM, and GradCAM++ for visual explainability to identify predictive features and areas of interest.

Awarded 'Best Undergraduate Capstone Project' by MIT-WPU School of CET and recognized for outstanding, practical research contributions

Achieved real-time biomedical waste detection and classification for a robot assisted system using YOLOv5 and Single Shot Detector (92.47% and 86.04% accuracy).

Built a visual dataset by collecting and manually annotating over 10,000 images of biomedical waste with the help of web scraping.

Created a 94% accurate computer vision model to aid farmers in identifying dragon fruit diseases and gauging maturity.

Created a custom dragon fruit image dataset with 2563 images— a first of its kind.

Implemented YOLOv5, Region-based Convolutional Neural Networks (RCNN), Inception v3, ResNet50, VGGNet, and Mask RCNN.

Built a web app to help users check if their medications are safe to take together and understand potential risks. Developed using React, Flask, and MongoDB, and powered by a structured drug interaction database. Integrated the OpenAI API to generate AI-based explanations of interactions. Designed a graph visualization of interactions with nodes and edges, added risk levels at a glance, and built a detailed table with explanations for each drug pair to support informed decision-making.

Developed a RESTful Web API in C#, ASP.NET Core, and .NET 6 for Pokémon management, enabling full CRUD operations through HTTP verbs.

Created a data access layer with Entity Framework Core for database connectivity and data retrieval.

Constructed and managed a SQL database using Azure Data Studio to store review data efficiently.

Developed an Event Search web app with Angular, TypeScript, and Bootstrap, & an iOS app with Swift, following the MVC design pattern.

Engineered a responsive backend with Node.js/Express and deployed the entire solution on Google Cloud Platform.

Integrated Ticketmaster, Spotify, Facebook, Twitter, Google Maps Geocoding, ipinfo.io APIs to provide event, artist, and venue information.

Enhanced user experience with event clicks, seamless navigation, autocomplete, location detection, social sharing, and favoriting features.

Created a custom Go-playing agent on a 5x5 board, employing Minimax with Alpha-Beta Pruning (from scratch), utilizing only NumPy.

Competed against some basic as well as more advanced AI agents such as:
Random Player (Moves randomly)
Greedy Player (Places the stone that captures the maximum number of enemy stones)
Aggressive Player (Looks at the next two possible moves and tries to capture the maximum number of enemy stones)
Alpha-beta Player (Uses the Minimax algorithm with alpha-beta pruning)
QLearning Player (Uses Q-Learning to learn Q values from practice games and make moves intelligently under different game conditions)

Achieved perfect records against Random, Greedy, Aggressive, & Alpha-Beta agents with 10/10 wins, and secured 5 wins against Q-Learning.

Implemented basic and memory efficient Dynamic Programming algorithms for the DNA Sequence Alignment of two strings.

Computed the cost of the alignment, first string alignment, second string alignment (Consisting of A, C, T, G, _ (gap) characters), time and memory usage of the two solutions.

Visualized and analyzed the CPU time and memory usage of both the solutions by generating plots of the CPU time vs problem size and of Memory usage vs problem size for the two solutions.

In this problem, I built a classifier that distinguishes images of Martian terrain with frost.

This dataset was created to study Mars’ seasonal frost cycle and its role in the planet’s climate and surface evolution over the past 2 billion years. The data helps in identifying low-latitude frosted microclimates and their impact on climate.

An interesting task in machine learning is classification of time series. Here, I have classified the activities of humans based on time series obtained by a Wireless Sensor Network.

The dataset comprises seven folders, each corresponding to a different type of activity. Within these folders are numerous files, where each file captures a moment of a person engaging in one of the activities. Every file consists of six time series, all of which are derived from the same individual's activities. The dataset encompasses 88 such instances, with each instance containing six time series. Furthermore, each time series consists of 480 sequential values.

Addressed missing data to preserve valuable information.

Calculated variability to identify key features.

Improved model accuracy by addressing data imbalance.

Adopted XGBoost for nuanced modeling.

Enhanced performance with SMOTE for class balance.

Implemented a MLP from scratch using no machine learning libraries.

Constructed and trained the neural network classifier using provided labeled training data from four different datasets: Spiral, Gaussian, XOR, Circle.