Shreyas Grampurohit

Dual Degree Thesis | Guide: Prof. Ajit Rajwade & Prof. Satish Mulleti

• Designed a noise-robust Deep Learning solver for Quantitative Group Testing, achieving reconstruction performance matching state-of-the-art combinatorial algorithms
• Established structural verifiability by introducing novel Jacobian analysis, showcasing that the network implicitly internalizes the combinatorial structure of the problem
• Demonstrated superior signal recovery resilience in regimes with sparse measurement noise compared to model-based solvers like Approximate Message Passing
• Extending the framework to real-valued signal recovery for applications like viral load estimation with rigorous theoretical guarantees

In-semester Research | Guide: Prof. Vijay Janapa Reddi, Harvard University

• Spearheaded "Archipedia," a one-billion-token dataset covering 50 years of premier computer architecture research, leading a team of 10 students
• Co-authored QuArch benchmark for evaluating LLMs in computer architecture; paper accepted to IEEE CAL and selected for 'Best of CAL' at HPCA 2026
• Architected metadata extraction and validation pipelines on AWS to parse thousands of academic manuscripts

Research Project | Guide: Prof. Ajit Rajwade & Prof. Satish Mulleti

• Developed a robust reconstruction framework for Cryo-EM Tomography addressing the ill-posed inverse problem of recovering 2D images from corrupted projections
• Formulated a three-way alternating minimization algorithm with shift-aware modified Graph Laplacian initialization
• Proved theoretical non-feasibility of spectral invariant estimators for arbitrary translational shifts
• Achieved 2.36× lower RRMSE compared to baselines; Best Poster Award at RISC 2025

Research Internship | Guide: Prof. Vijay Janapa Reddi, Harvard University

• Contributed to ArchGym, an open-source framework connecting diverse ML algorithms to architecture simulators
• Developed ML-based proxy models to mitigate simulator latency, optimizing the accuracy-speed trade-off
• Led automated documentation generation and authored comprehensive guides for research reproducibility