__global__ void greeting() {

Shubhodeep Mitra.

device void buildScalableSystems() { ... }

Hi, I’m a curious engineer driven by a deep passion for solving complex challenges at scale. I love building at the intersection of Distributed Systems, HPC, MLE, and AI Infrastructure.
Whether it’s architecting resilient systems, optimizing for performance, or enabling scalable ML training, I thrive in environments where engineering meets innovation.
Currently, I'm focused on building resilient financial systems at Marqeta.

Get In Touch

host class About {

Hello! I'm Shubhodeep, a software developer with a passion for building scalable and resilient infrastructure. My journey in software development began with a fascination for Distributed Systems, High-Performance Computing, and AI Infrastructure.

I hold a Master of Science in Computer Science with a thesis from Arizona State University, where I focused on Distributed Systems and AI/ML Systems. I'm open to SDE, MLE, and AI Infrastructure roles, aiming to apply my experience to develop innovative solutions for real-life problems.

Here are a few technologies I've been working with recently:

__constant__ char* skills[] = {
"Java",
"C/C++",
"Python",
"Go",
"CUDA",
"PyTorch",
"Parallel Programming",
"AWS",
"Kubernetes",
"Docker",
"Spring Boot",
"MongoDB",
"PostgreSQL",
"Redis",
"LLM",
"RAG"};

global void getExperience() {

// November 2024 - Present

Software Developer 2

Marqeta, Bay Area, USA

Building scalable and available infrastructure for network systems to support mission-critical financial systems.
Collaborating cross-functionally to migrate load-balancing infrastructure from F5 to Envoy Proxy.

// December 2022 - July 2024

Graduate Research Assistant (Distributed Systems)

EMITLab ASU, Tempe, USA

Led developments of scalable ML infrastructure to support scaling of large machine learning models, optimizing data pipelines and concurrent workflows on AWS, which increased model diversity and training efficiency by 4x.
Improved parallel processing and automated ML workload deployments using Ansible on AWS EC2 clusters, reducing experiment runtime by 27% through improved parallelism and resource utilization.
Revamped in-memory graph data structures and integrated local caching mechanisms, reducing MongoDB calls by over 50,000 and speeding up generation of provenance graphs.
Built an interactive data visualization platform with React.js and Next.js to summarize and analyze multi-variate time series, aiding data-driven decision-making for large-scale datasets.

// July 2018 - July 2022

Software Engineer 2

Hewlett Packard Enterprise (HPE), Bangalore, India

Led development efforts in C/C++ for critical networking components, including internal-VLAN, L3 counters, Netdev, and Ofproto, pivotal for IP-Subinterface, facilitating seamless traffic flow across 17 protocols.
Redesigned multicast protocols, IGMP and MLD to integrate real-time packet flow monitoring, boosting reliability.
Developed CLI infrastructure for L2 protocols, VLAN Translation and Multi-Zone User-Based Tunneling, integrating SDN features that enabled network management and simplified configuration for large-scale deployments.
Engineered test automation suite for all development work, achieving 98% code coverage and empowering DevOps team to perform continuous system health checks.

device struct Projects[] = {

AI Agent for Personalized K–12 Tutoring

Built an AI agent combining Gemini, Tavus AI, and a RAG pipeline to deliver 1-on-1 human-like video tutoring with adaptive feedback, retrieving curriculum-aligned content to enable dynamic, personalized learning journeys.

LLM RAG Gemini

ColumnarDB

Designed a Columnar Database system in Java with features including BitMap and BTree indexing, compressed BitMap, Columnar Joins, Columnar Sort, Scan, and Delete optimizing data operations efficiency for 50k entries.

Java Databases Indexing

Image Retrieval and Recommendation

Designed and built an end-to-end image retrieval and recommendation system with 92% accuracy, leveraging vector embeddings from neural network features, dimensionality reduction, Personalized PageRank, and LSH.

Vector Embeddings PageRank LSH

Elastic Face Recognition Service

Developed a scalable face recognition service using AWS EC2, S3, DynamoDB, and SQS, achieving 98% accuracy in facial recognition from video streams while dynamically scaling to handle 10,000+ concurrent requests.

AWS DynamoDB SQS

Scalable Aesthetic-Preserving Face De-Identification

Created a Kotlin Android app employing an ML-kit and openCV foundation to detect individuals in photograph backgrounds through face recognition, applying an aesthetic-preserving filter to safeguard bystanders' privacy.

Kotlin ML-kit OpenCV

Live Video Streaming Application

Architected and implemented a high-performance live video streaming application in Go and WebRTC, optimizing for low latency and scalability, employing advanced techniques to enhance video delivery quality.

Go WebRTC Streaming

#include "research.cuh"

Pancommunity: Non-Monolithic Complex Epidemic and Pandemic Modeling

SSRN, 2024

The PanCommunity project investigates the impact of testing, preventative measures, and vaccines on community response and resilience during epidemics. It develops a platform that supports seamless integration of independently developed component models to analyze various environmental and behavioral factors for improving pandemic response.

PDF

DataStorm-EM: Exploration of Alternative Timelines within Continuous-Coupled Simulation Ensembles

arXiv, 2024

This paper presents the DataStorm-EM platform for data and model-driven simulation ensemble management, optimization, analysis, and exploration. It addresses challenges in generating and exploring alternative timelines from continuous-coupled simulation ensembles to support decision-making in complex socio-economical domains.

PDF

Visual Summarization of Multi-Variate Time Series

ProQuest, 2024

This research focuses on developing methods for visually summarizing multi-variate time series data, enabling more effective analysis and interpretation of complex temporal datasets across various domains.

PDF

Novel TLS Signature Extraction for Malware Detection

IEEE Conference, 2020

This paper presents a solution to identify the presence of malware in a network flow from the initial unencrypted Client Hello packet of TLS handshake. We perform feature engineering on the subfields of TLS metadata to extract interpretable signatures for binary classification.

PDF

Smart Wheelchair using IOV

IJCEA, 2018

This paper aims to develop a wheelchair that can move autonomously using the Internet of Vehicles (IOV). The proposed model uses IOV and image processing to help the wheelchair move autonomously, enabling integration with smart city architectures.

PDF