cv

Software Engineer

Tangoe Inc. (contractor at Jeavio)

Developed scalable backend systems and optimized financial data processing workflows, improving efficiency and automation in invoice management and reconciliation.

• Optimized invoice processing by developing an Invoice Exception Processing (IEP) system, consolidating data from five sources, handling 2M+ invoices/month, and boosting efficiency by 30%.
• Enhanced PostgreSQL performance, reducing data retrieval times by 35%, improving responsiveness for high-volume financial transactions.
• Built a scalable Java backend with RESTful APIs (Spring Boot), improving data handling speeds by 40% and streamlining inventory reconciliation.
• Developed an invoice classification model using self-supervised graph-based neural networks, accelerating decision-making for analysts.
• Automated exception resolution with Pentaho ETL pipelines & Flowable workflows, cutting analyst workload by 50% and handling 500M Kafka messages/month.
• Designed a GPT-2-based mail service to auto-generate custom alerts, ensuring rapid response to critical exceptions.
• Achieved 95% code coverage with rigorous unit testing (JUnit, Mockito, Jasmine, Karma), ensuring software reliability in production.

Software Engineering Intern

Tangoe Inc. (contractor at Jeavio)

Engineered RESTful APIs and scalable Java backend architectures. Built custom Gantt charts with Angular and Node.js for resource management.

• Developed a Resource Management Application for HR and managers, streamlining resource allocation across teams and projects—a solution still in active use.
• Built custom Gantt charts in Angular with a scalable Node.js backend, enabling secure, multi-user access with end-to-end authentication. Designed a flexible data architecture using GraphQL and MongoDB, integrating NoSQL and graph features for efficient cross-team tracking.
• Achieved 70% code coverage with Jasmine & Karma, ensuring software reliability through rigorous testing.

Head

Computer Society of India (CSI)

Engineered RESTful APIs, built scalable backend systems, and developed interactive data visualization tools for resource management.

• Led a team of 20+ members, organizing workshops and events to enhance technical skills and promote knowledge sharing.
• Developed a web-based platform for managing resources and events, improving efficiency and collaboration within the organization.
• Implemented a data visualization tool using Tableau, enabling real-time tracking of budget allocation and event participation.
• Collaborated with local colleges to organize hackathons and coding competitions, fostering a culture of innovation and creativity among students.

Master of Science

New York University

Computer Science

• Machine Learning
• Neuroinformatics
• Design and Analysis of Algorithms
• Principles of Database Systems
• Computer Vision
• Artificial Intelligence
• Network Security
• Master's Thesis

Bachelor of Engineering

Gujarat Technological University

Computer Engineering

• Compiler Design
• Information Security
• Software Engineering
• Operating Systems
• Cloud Computing
• Math - I, II, III, IV

Winner of INDRA-9 2019

India's Next Development by Renewable Energy and Astronomy

Created an AI-based system that would utilize the already optimized drip irrigation and enhanced it by using crop data and weather data to predict the water requirement of the crop in an online system.

Self-Supervised Graph-Based BrainAGE Model

Developed a self-supervised graph neural network for multiple downstream tasks using MRI-derived structural features. Designed a 148-node graph representation using the Destrieux Atlas.

• Developed a self-supervised graph-based model for multiple downstream tasks using MRI-derived structural features.
• Integrated the ABCD dataset for training, employing data preprocessing and feature extraction techniques using the Freesurfer Software Suite.
• Created a 148-node graph representation using the Destrieux Atlas, mapping brain regions for connectome analysis.
• Implemented a self-supervised contrastive learning GNN encoder to learn brain representations without labeled data.

Brain Region Predictions Using LFP Data

Built a real-time classification pipeline for brain region localization using electrophysiological LFP recordings. Leveraged deep learning models like SimCLR and BrainBERT. Applied a wave2vec2 transformer for feature tokenization as an input to a classification-projection layer.

• Designed and implemented a real-time classification pipeline for brain region localization using LFP recordings.
• Leveraged SimCLR and BrainBERT deep learning models (to act as baselines and) to capture complex neural patterns in LFP data.
• Used wave2vec2 transformer for tokenizing temporal features, enabling high-quality input to a classification layer.
• Optimized the model for real-time inference, enabling continuous classification of neural data for neuroscience research.
• Conducted extensive model benchmarking, trying to achieve state-of-the-art performance in brain region prediction (Project still ongoing in the fine-tuning stages).

Subtyping and Staging MS Patients

Developed a disease progression model for Multiple Sclerosis using probabilistic modeling techniques - SuStaIn + TEBM. Collaborated with experts in MS research.

• Developed a disease progression model for Multiple Sclerosis (MS) using probabilistic models SuStaIn and TEBM.
• Utilized SuStaIn for subtyping and staging MS patients based on longitudinal clinical data.
• Collaborated with Peter Wijeratne to refine the model, providing valuable insights into disease progression.
• Applied probabilistic graph models to track MS disease stages, enabling early-stage diagnosis and intervention.
• Integrated advanced time-series analysis to predict patient outcomes and optimize treatment strategies.

Neurodegenerative Disease Detection via Enhanced EEGNet

Developed an enhanced EEGNet model to classify Alzheimer's, Mild Cognitive Impairment (MCI), and Frontotemporal Dementia (FTD) using EEG signals. Applied advanced preprocessing techniques.

• Developed a deep learning-based model (EEGNet) to classify neurodegenerative diseases from EEG signals.
• Enhanced EEGNet architecture to mitigate fast convergence issues and improve classification accuracy.
• Applied advanced EEG preprocessing techniques, including ICA artifact removal and ASR denoising, to improve signal quality.
• Achieved high classification accuracy for Alzheimer's, MCI, and FTD, supporting early diagnosis and intervention.
• Utilized EEGLab to preprocess and clean EEG signals before feeding them into the enhanced EEGNet model.

3D Point Cloud Generation from 2D Images

Built a pipeline to reconstruct 3D point clouds from 2D images using camera calibration, homography matrices, and vocabulary trees, without deep learning.

• Developed a pipeline to convert 2D images into 3D point clouds using geometric transformations and camera calibration.
• Implemented homography matrices and camera focal length estimation to create accurate 3D reconstructions.
• Leveraged vocabulary trees to match image features across multiple views, enhancing point cloud precision.
• Used Open3D for visualizing 3D point clouds and converting them into meshes for real-world object integration.
• Built a non-deep learning-based system for real-time 3D reconstruction from multi-view images.

Foreign Whispers

A web-based platform that uses AI to translate and download YouTube videos, leveraging Hugging Face models for high-quality translations.

• Built a web-based platform for translating YouTube videos using Hugging Face models for high-quality AI-powered translations.
• Developed the frontend using React to create a dynamic user interface for translating and downloading videos.
• Used Node.js to build the backend, handling video processing and integrating with the Hugging Face API for translation.
• Implemented features like automatic subtitles and multilingual support for seamless video translation.
• Optimized performance for faster translation processing, enabling users to translate videos quickly.