Introduction
I am an incoming PhD student (Fall 2026) in the Department of Electrical and Computer Engineering at The University of Texas at Austin, TX, USA, under the DICE track, focused on efficient and robust machine learning systems.
I worked as a Software Engineer at Deutsche Bank (Risk, Finance & Treasury) through July 2026, in the Platform Engineering team, focusing on resiliency of GCP and private cloud infrastructure, along with agentic development for building reliable systems.
I completed my B.E. in Computer Science and Engineering from Jadavpur University, Kolkata, India in 2024. During my undergraduate years, I worked as a researcher at the CMATER Lab under Prof. Ram Sarkar, focusing on deep learning for medical image analysis, attention mechanisms, and model compression. This work resulted in publications at MICCAI 2025 (first author), ICONIP 2024, and IEEE Transactions on Artificial Intelligence (first author).
My current research area is primarily efficiency and robustness in large-scale machine learning systems under Prof. Michael Orshansky. My prior research experience includes representation learning, network pruning in Vision Transformers, adversarial robustness under class imbalance, and resource-efficient multimodal learning. I have been an active hackathon participant, with top finishes in nine national and international competitions, including wins at Nx EVOS, LabLab.AI, and DevRev.
Academic & Professional Timeline
PhD ECE – UT Austin
Incoming PhD student focused on efficiency and robustness in large-scale machine learning systems under Prof. Michael Orshansky.
Software Engineer – Deutsche Bank
Cloud infrastructure, DevOps, LLM-based NLP pipelines, Agentic Platform for microservice failure analysis.
Summer Intern – Deutsche Bank
Built end-to-end chatbot with Flask & React on noisy internal data.
Jadavpur University
B.E. in Computer Science & Engineering. Undergraduate researcher at CMATER Lab under Prof. Ram Sarkar.
Cooch Behar Rambhola High School (WBCHSE)
Completed Grade 12 board exams. Qualified WBJEE 2020.
Cooch Behar Rambhola High School (WBBSE)
Completed Grade 10 board exams.
Research & Publications
Current Focus: I currently focus on efficiency and robustness in large-scale machine learning systems, including distributed training and optimization for large-scale models (VLM, LLM).
Past Experience: representation learning, model compression, attention-based and Vision-Language architectures, self-supervised learning, robustness under data imbalance, redundancy analysis in CNNs/Transformers, and graph-based attention mechanisms for medical imaging.
Publications & Preprints
Background-Invariant Independence-Guided Multi-head Attention Network for Skin Lesion Classification First Author
Proposes BIIGMA-Net to address visual artifacts, illumination bias, and class imbalance in dermatological imaging. Introduces Multi-head Independence-Guided Channel Attention (MICA) using HSIC to minimize redundancy among attention heads, and a saliency-driven background suppression module to enforce lesion-centric feature learning. Validated on ISIC-17/18/19, beating SOTA F1 on ISIC-18/19.
Advisors: Prof. Friedhelm Schwenker, Prof. Ram Sarkar
Paper Link
Multi-headed Graph-Based Attention Aided U-Net for Nuclei Segmentation
Extends U-Net with a Pairwise Node Similarity Attention Module (PNSAM) to capture long-range dependencies via kernel-based pairwise relationships. Incorporates multi-headed attention and channel pruning (50–70% zero-variance channels removed) to improve feature diversity and calibrate spatial attention scores. Tested on Monuseg, TNBC, and CryoNuSeg, outperforming prior methods by 1.0–2.5% IoU.
Advisors: Prof. Dmitry Kaplun, Prof. Ram Sarkar
Paper Link
Margin-Aware Adaptive-Weighted Loss for Imbalanced Classification First Author
Develops MAAW Loss integrating margin-based softmax with adaptive reweighting to stabilize training under extreme class imbalance. Quantifies prediction uncertainty using confidence-driven minibatch statistics to improve generalization. Validated on CIFAR-10/100, FMNIST (imbalance up to 1:100), and clinical datasets HAM10000 and APTOS-2019 BD, showing significant gains in Macro F1.
Advisor: Prof. Ram Sarkar
Paper Link
Research Philosophy
Core Theme: Trustworthy AI—Unifying Efficiency, Robustness, and Interpretability.
- Move beyond ‘accuracy’ to ‘reliability’ in OOD scenarios.
- Bridge the gap between theoretical robustness and deployment constraints (Industry Alignment).
Deep technical rigor over ‘vibe coding’—committed to first-principles understanding.
Honors & Achievements
- Invited attendee at IndoML Symposium 2024, IIT Bombay.
Education
The University of Texas at Austin, TX, USA
Fall 2026 – 2031 (Expected)PhD in Electrical and Computer Engineering (Decision, Information, and Communications Engineering - DICE Track)
- Incoming PhD student in the Department of Electrical and Computer Engineering.
- Research focus: efficiency and robustness in large-scale machine learning systems under Prof. Michael Orshansky.
- Recipient of the UT Austin Engineering Fellowship (multi-year).
Jadavpur University, Kolkata
2020 – 2024B.E. in Computer Science & Engineering — 9.23 CGPA (9.30 SGPA) — First Class Distinction with Honours
- Secured 4th(/95) merit position in the department. 📄 Transcript
- Worked at CMATER Lab under Prof. Ram Sarkar as an undergraduate researcher, focusing on Deep Learning and Computer Vision.
- Core coursework: OS, DBMS, AI, NLP. Honours: Machine Learning, Big Data Analytics, Network Security, Bioinformatics, Optimization Techniques.
Cooch Behar Rambhola High School, West Bengal
Up to 2020WBBSE & WBCHSE Board Exams
- Grade 10 (2018): Scored 97.4%, ranked 8th in the board (~1M students).
- Grade 12 (2020): Scored 97.2%, ranked 13th in the board (~760k students).
- WBJEE 2020: Rank 139th (Open Category) among 75k candidates.
Career
Deutsche India Pvt. Ltd. — Risk, Finance & Treasury
July 2024 – Jul 2026Software Engineer
- Worked on Cloud infrastructure and DevOps using GCP, Kubernetes, CI/CD pipelines, and SRE tools to maintain high-availability financial systems.
- Designed and implemented LLM-based NLP pipelines for derivatives analysis, mapping quantitative market factors to financial news for explainable insights.
- Developed Agentic Platform to perform root-cause analysis of microservice failures, improving system observability and reducing incident resolution time.
- Optimized large-scale model deployment through architecture tuning, quantization, and retrieval strategies.
Deutsche India Pvt. Ltd. — Technology, Data & Innovation
Summer 2023Summer Intern
- Built an end-to-end chatbot with Flask (backend) and React (frontend) on extremely noisy and unstructured internal data.
- Implemented context search algorithms and dynamic knowledge base expansion on language models running with limited resources.
Contests & Hackathons
I actively participate in contests and hackathons to immerse myself in diverse challenges, interact with industry professionals, and collaborate with people from different backgrounds. To date, I have secured top positions in nine of them across both international and national stages, yielding $8,500+ prize money in total.
🥉 2nd Runner Up — Hacknosis 2024: Sustainable MedTech Hackathon
Developed CarbonSage: a comprehensive solution for calculating, tracking, and reducing carbon emissions from devices and data usage using a Multi-Agent System for snippet-by-snippet emission count and code analysis.
Hackathon Link
🏆 Winner — Nx EVOS 2024: Enterprise Scale Video Application
Built an enterprise-scale video object detection pipeline for retail. Processes CCTV feeds in real-time, computes product dwell times, tracks purchases, and uses the Apriori algorithm for optimal product placement. A Multimodal System involving Signal-to-NLP Transformation.
Hackathon Link
🥇 Best Project by LlamaIndex — Advanced RAG Hackathon (LabLab.AI)
Developed an industry-grade CRM platform that ingests product reviews, breaks them into trigger phrases using smaller LLMs, and routes through a hierarchical retrieval system to generate actionable tickets.
Project Link
🏆 Winner — KnackToHack by M&G Global Services
Developed a Safe Guard for LLM chatbots that automates rule creation from regulations like the EU AI Act and GDPR, forming a knowledge graph for hierarchical RAG to ensure compliance.
Proof
🥇 First Place — DevRev Forge 2024 by DevRev
Built TicketIn'site.AI: one-stop solution for comprehensive analysis of app reviews and user feedback from diverse sources, facilitating comparative trend analysis of various insight factors using AI.
First Place, DevRev Forge by DevRev, 2024 National.
Project Link
🥉 2nd Runner Up — App-A-Thon 2.0 by Awiros & Intel
Developed a real-time rider safety application using pose detection, deployed with MLOps on AWIROS, the world's first Computer Vision OS.
Proof
🏆 Winner — DataFest @ Integration 2023 by ISI Kolkata
Won a Kaggle contest with two challenges: a signal processing task to identify whale sounds and an estimation task to count tigers from paw prints.
Credential
Miscellaneous
Skills & Certifications
Languages
C C++ Java Python JavaScript MATLAB ASM VHDL
Software & Tools
TensorFlow PyTorch CUDA OpenCV LangChain Docker Git Flask AWS
Core CS & Honours
OS Networks DBMS AI NLP Machine Learning Big Data Network Security Bioinformatics
Certifications
Hobbies & Interests
Playing piano and listening to music. Trained in Indian Classical music and Rabindra Sangeet (still keeping it up).
I love reading novels, especially retro Bengali classics 😊, though I have not been able to keep up as consistently in the last few years 😔.
Playing games on PS.