cv

ML/Data Engineering Summer Intern - Mailchimp Core Data Products Team

Intuit

ML/Data Engineering Intern on Mailchimp's Core Data Platform

• Developing large-scale ELT pipelines and data workflows using PySpark, Apache Airflow (workflow orchestration), BigQuery, Dataform, Terraform (infrastructure as code), Kubernetes, and GCP for Mailchimp's Core Data Platform.
• Designing an agentic MCP-server-driven system to automate the CI/CD validation stage, stage-to-production data-diff analysis, anomaly detection, and deployment verification, improving release reliability and reducing manual effort.

Graduate Teaching Assistant - Data & Visual Analytics (CSE 6242)

Georgia Institute of Technology

• Supporting 1500+ graduate students globally in the Data & Visual Analytics course.

Graduate Student Researcher

Internet Intelligence Lab, Georgia Tech

Research on RPKI and BGP/network security

• Working on RPKI (Resource Public Key Infrastructure) and network security research under the Special Problems in Networking program.

Software Engineer - Assistant Vice President

Wells Fargo

Software Engineer in Finance Technology (Corporate Treasury - Valuation & Product Control) LoB, Independent Price Verification (IPV) Team

• Led the end-to-end development of the IPV web application, supporting accurate pricing & valuation of financial portfolios worth $200B, following Agile and SCRUM methodologies, CI/CD, JIRA, and SDLC best practices.
• Designed and delivered production-grade, end-to-end tested full-stack features using Django (backend services), Python and Pandas (ETL), Angular (frontend), SQL Server (relational database), Jenkins (CI/CD), and pytest for unit testing.
• Architected a large-scale Python-based ETL data pipeline for data ingestion and processing of 20M+ records daily from 130 sources including database extractions, FTP servers, APIs, and file reads.
• Optimized query performance by 66% (10s to 3s) through indexing and archival processes created using stored procedures.
• Automated generation and validation of 50+ commodity control sheets using Python scripts, eliminating 250,000+ manual formula operations and saving 8 hours of effort, resulting in a 99.6% reduction in processing time.
• Awarded the Team Spotlight Award for improving application availability and system reliability by 46%, reducing 100+ recurring yearly production issues, and enhancing monitoring through automated alerting pipelines.

Research Intern - Multimodal LLMs, Machine Learning, Vision-Language Models

Multimodal Digital Media Analysis (MIDAS) Lab, IIIT Delhi

• Evaluated the reasoning capabilities of 5 Multimodal Large Language Models in the geometry domain, including LLaVA, G-LLaVA, and Gemini Pro Vision, using chain-of-thought, zero-shot, few-shot, and fine-tuning methodologies.
• Led a team of 3 to create GeoVQA, a high-quality Indian-context geometry benchmark dataset with 4K images; conducted experiments across GeoVQA, PGPS9K, and Geometry3K to assess reasoning performance.
• Devised the Mixture of Refinement Agents (MoRA) agentic framework to iteratively reduce reasoning errors in physics tasks, improving accuracy by up to 16% on SciEval and MMLU for open-source LLMs (LLaMA-3, Gemma-2). Accepted at AAAI'26 Workshop.

Software Engineering Intern - Treasury and Trade Solutions (TTS) Unit

Citi

Summer Intern in Treasury and Trade Solutions (TTS) Business Unit

• Proposed and migrated the UI of an in-house Process Automation Tool from Thymeleaf to Angular, used for productivity tracking and vendor resource management in the Payments and Receivables unit.

Attitude Determination and Control Subsystem (ADCS) Contributor

ISRO COEP Satellite Team Initiative (CSAT)

• Implemented the TRIAD algorithm and an Unscented Kalman Filter in C to determine satellite attitude and improve the accuracy of attitude estimation, respectively, with testing performed using real satellite data.

Master of Science in Computer Science (MSCS)

Georgia Institute of Technology

Computer Science - Machine Learning Specialization

• Machine Learning
• High Performance Computing
• Natural Language Processing
• Deep Learning
• Data & Visual Analytics
• Special Problems in Networking (RPKI & BGP Security)

Bachelor of Technology (B.Tech), Honors in Data Science

COEP Technological University (Formerly College of Engineering Pune)

Computer Engineering

• Operating Systems
• Networking
• Databases
• Data Structures and Algorithms
• Software Engineering
• DevOps
• Web Technologies
• Object-Oriented Design
• Cloud Computing
• Network Security
• System Administration
• Big Data Analytics

Institute Gold Medalist and Valedictorian

COEP Technological University

Gold medal awarded for being Computer Engineering branch and Institute topper (Rank 1/780) with a CGPA of 9.81/10

Best Performer of Mathematics at UG Level

Alumni Association of COEP

Awarded for scoring a perfect grade (10/10) in all Mathematics courses (Linear Algebra, Calculus, Differential Equations, Probability & Statistics) at undergraduate level.

Certificate of Merit (Third Year Topper)

Alumni Association of COEP

Awarded for 1st Rank in T.Y.B.Tech (Computer Engineering) with CGPA of 9.86/10.

Frank Hirschvogel Excellence Award

Frank Hirschvogel Foundation

Award presented for the 1st Rank in Institute and Computer Engineering Department in T.Y.B.Tech with CGPA of 9.81/10.

Certificate of Merit (Second Year Topper)

Alumni Association of COEP

Awarded for the 1st Rank in S.Y.B.Tech (Computer Engineering) with a CGPA of 9.89/10.

Improving Physics Reasoning in Large Language Models Using Mixture of Refinement Agents

Accepted at AAAI 2026 Workshop (arXiv:2412.00821)

GeoVQA: A Comprehensive Multimodal Geometry Dataset for Secondary Education

IEEE 7th International Conference on Multimedia Information Processing and Retrieval (MIPR)

A Comprehensive Survey of Weed Detection Methodologies in Soybean Crop

IEEE 3rd Asian Conference on Innovation in Technology (ASIANCON)

Attitude determination using a system of sensors and UKF for a solar sailing nanosatellite

International Astronautical Federation

MPI Communication Primitives & Parallel Monte Carlo Simulation - High Performance Computing

• Implemented a parallel Monte Carlo simulation to estimate pi using MPI on Georgia Tech PACE HPC cluster infrastructure, distributing workloads across processors and achieving scalable performance up to 10^9 samples.
• Developed custom MPI collective communication primitives (Scatter, AllGather, AllReduce, AlltoAll) using MPI_Send/MPI_Recv and executed experiments using SLURM (sbatch, srun) to analyze communication overhead and parallel scalability.

Shuttle-Route Mobility Analytics Platform (NYC Taxi Data) - Data & Visual Analytics

• Built a distributed PySpark pipeline on Databricks to clean, ingest, and preprocess parquet-based multi-year NYC TLC data (50M+ trips), performed temporal feature engineering and cluster-to-cluster origin-destination aggregation.
• Applied weighted K-Means and DBSCAN to uncover dense mobility hubs and high-frequency travel corridors.
• Developed an interactive geospatial analytics system using Streamlit, integrating Folium, GeoPandas, Plotly Sankey, and 3D Deck.gl layers to visualize pickup densities and shuttle-route flows, backed by cached Spark tables for low-latency exploration.

Robust AI-Generated Text Detection under Decoding Shift - Natural Language Processing

• Performed a thorough ablation study of AI-generated text detection under decoding shifts (temperature, top-p, greedy, beam search) and back-translation paraphrase attack, with AUROC, TPR@1%FPR, and log-perplexity as evaluation metrics.
• Benchmarked zero-shot Binoculars and supervised detectors (Kim-CNN, RoBERTa) under varied generation and sampling strategies.
• Constructed a balanced 20K-sample subset of TuringBench to ensure fair robustness evaluation under distribution shift.

Superpixel-Based Semantic Segmentation using Vision Transformers - Machine Learning

• Developed an efficient semantic segmentation pipeline following the full ML lifecycle - Cityscapes data preprocessing, feature engineering, and ViT-based model training - achieving 53.77% mIoU, 90.74% pixel accuracy, and ~55 ms inference latency per image.
• Implemented a CUDA-accelerated SLIC module and integrated ResNet-50 hypercolumn features with a Vision Transformer (ViT) encoder, reducing pixel-level computation from 2M to 2K tokens (1000x reduction).

Generative AI-Driven Sustainability Benchmarking Automation Tool

• Spearheaded the design and deployment of a full-stack cloud and LLM-based web application on Azure App Service during a company-wide hackathon; led a team of 4 to automate Environmental, Social, and Governance (ESG) PDF report analysis.
• Engineered a RAG pipeline by integrating Azure Document Intelligence for PDF-to-text extraction, Azure Search Index for embeddings storage & context retrieval, and Azure OpenAI APIs for Q&A on 100+ page ESG documents.

Weed Detection using Vision Transformers (ViT) - Final Year B.Tech Thesis Project

• Formulated the weed detection problem as 2 supervised learning tasks - classification and object localization/detection - proposing a ViT as backbone for Detection Transformer (DETR), applying regularization techniques to mitigate overfitting.
• Achieved mAP@0.5 of 82.9% on the CottonWeed dataset (7% improvement over CNN-based detectors).
• Published a survey paper at IEEE ASIANCON, providing a detailed taxonomy of weed detection techniques and the future potential of ViT in real-time smart farming solutions.

User-Threads (threadHive) - Operating Systems

• Built a user-level multithreading library in C, supporting 3 mapping models: one-one, many-one, many-many.
• Exposed APIs for thread lifecycle control and management, implemented synchronization primitives like spinlock and signal handlers for over 50+ signals.
• Validated models using race conditions and achieved 80% reduction in execution time for matrix multiplications.

HTTP Server (HTTPHaven) - Computer Networks

• Developed a multithreaded HTTP/1.1-compliant apache-like web server in Python following RFC 2616 standards.
• Implemented 5 HTTP methods (GET, POST, PUT, DELETE, HEAD), 15 status codes, 5 media types, and multipart form data handling.
• Designed a customizable configuration file (similar to apache.conf) for server-side settings with 12+ configurations (e.g., Keep-Alive, timeout, persistent connections).

BankTrack - Banking Application

• Built a full-stack banking application simulating account management, fund transfer, and admin operations, following a 1-sprint Agile workflow with 10+ integrated features.
• Implemented JWT-based authentication, email/OTP verification, and role-based REST APIs using Spring Security and React.

Paint Application - Object Oriented Programming

• Developed using 3 pillars of OOP (inheritance, encapsulation, polymorphism) in C++ with OpenGL for graphics and GLUT for event handling.
• Added 5+ features, including shape drawing algorithms, undo/redo, erase, area filling, border coloring, and brush tools.

Arbitrary-Precision Calculator - Data Structures & Algorithms

• Built a bc command-like calculator in C supporting arbitrary-precision arithmetic (up to 10,000+ digits), with user-defined data structures, doubly linked list digit storage, and sign/precision handling.
• Implemented a stack-based evaluator with a Finite State Machine-based parser to compute 50+ infix expressions/second, supporting 6 arithmetic operations.