Machine Learning Researcher

Shivi Gupta

******@******.***.*** • 341-***-**** • linkedin.com/in/shivigup/

EDUCATION

Carnegie Mellon University - GPA: 4.0/4.0 Pittsburgh, PA Master of Science in Electrical and Computer Engineering May 2025

• Courses: Computer Architecture, Parallel Computing, Distributed Optimization, Large Language Models, Visual Learning

• Graduate Teaching Assistant: Signals and Systems, Mathematical Foundations of Electrical Engineering Indian Institute of Technology Kanpur - GPA: 9.2/10.0 (Distinction) Kanpur, India Bachelor of Technology in Electrical Engineering May 2023

• Double Major: Physics, Minors: Computer Science - Machine Learning, English Literature

• Courses: Image Processing, Digital Signal Processing, Machine Learning for Signal Processing, Control Systems Analysis SKILLS

Languages & Frameworks: C/C++, Python, JavaScript, MATLAB, TensorFlow, Keras, PyTorch, OpenCV, Networkx Utilities: Git, Linux, LATEX, SQL, HTML/CSS, AWS, GCP, Docker, Azure, Gem5, Verilog, OpenGL, CUDA, Android Studio, Vulkan INTERNSHIP EXPERIENCE

Schlumberger (SLB) Cambridge, MA

Advanced Acoustics Intern May 2024 - Aug 2024

• Advanced the capabilities of Epilogue, an acoustic barrier evaluation tool by integrating heterogeneities and introducing an uncertainty metric to generate high resolution ultrasonic-like features from lower resolution sonic data

• Designed pipelines in a domain generalization and ordinal regression methodology with a user-friendly interface by shell scripts and Python for Kubernetes deployed on Git and Docker containers, resulting in a 4% improvement in accuracy Adobe Bangalore, India

Research Intern May 2021 - Aug 2021

• Collaborated with a team of four interns to migrate user profiles from cloud to edge servers, by leveraging sequence modeling, autoencoders, and multitask learning on user interaction data, producing low 16-dimensional universal vectors

• Filed a patent “Generating Concise and Common User Representations for Edge Systems from Event Sequence Data stored on Hub Systems” Adobe P10960-US as a co-inventor

RESEARCH

Electrical and Computer Engineering, Carnegie Mellon University Pittsburgh, PA Graduate Research Assistant Jan 2025 - Present

• Designed a biologically inspired Content-Addressable Memory (CAM) system for neuromorphic computing by implementing hierarchical reference frames and dynamic memory composition achieving a 16x memory footprint reduction

• Implemented a C++-based spatial learning mouse-in-a-maze simulator enabling representation of complex environments with sparse, composable data structures under constrained neuromorphic memory budgets Electrical Engineering, Indian Institute of Technology Kanpur Kanpur, India Undergraduate Researcher Jan 2021 - Apr 2021

• Developed an HTML experiment for data collection on color selection for emotion-inducing images, demonstrating a significant emotion-color association in humans with a correlation of 0.34 in pre-trained classification models

• Co-authored a paper “Emotion-Color Association in Biologically Inspired Deep Neural Networks” (Gupta, S., & Gupta, S. 2021), published in the Proceedings of the Annual Meeting of the Cognitive Science Society, 43rd edition PROJECTS

Parallelizing OpenCV GrabCut – K-Means and Weights Fast Code II, CMU Jan 2025 - May 2025

• Accelerated OpenCV’s GrabCut segmentation pipeline through CUDA based parallelization of K-Means and Graph weight calculation, optimizing memory access with coalesced loads and shared memory reuse, achieving a 120x speedup

• Redesigned GrabCut initialization and GMM setup by replacing OpenCV classes with GPU-friendly structs and implementing structure-of-arrays memory layout to enhance spatial locality and minimize kernel launch overhead Persuasiveness of Large Language Models Large Language Models, CMU Sep 2024 - Dec 2024

• Designed language models for rating persuasiveness by instruction tuning and fine-tuning open-sourced language models on Anthropic Persuasion dataset, achieving a Cohen’s Kappa of 0.3 for zero shot with human ratings

• Fine-tuned GPT models for generating persuasive arguments for given claims, resulting in a perplexity score of 2082 demonstrating complex and nuanced generations

Conditional Point Cloud Generation Visual Learning and Recognition, CMU Sep 2023 - Dec 2023

• Developed a model for generating point clouds from text prompts and Canny edge images using ControlNet integrated with a fine-tuned stable diffusion framework, achieving P-IS scores comparable to leading single modality baselines

• Optimized Stable Diffusion by implementing low-rank adaptation resulting in cleaner intermediate 2D images, feeding them to a Wonder-3D model to generate 3D meshes, subsequently sampled into point clouds

Contact this candidate