May 2025 – Sep 2025 · 4 mos · Ann Arbor, Michigan, United States

• Project: ML based Sequential Change Detection
• ➔ Co-developed ML-based change detection algorithms that generalize the CuSum method using f-divergence-driven likelihood ratio estimation, achieving near-optimal detection delay with minimal distributional assumptions.
• ➔ Designed a novel classifier-based sequential detection framework for multi-class scenarios using e-processes and betting-based tests, enabling robust changepoint detection without density estimation.
• ➔ Led empirical benchmarking of model architectures, divergence measures, and hyperparameters, uncovering optimal design choices and identifying limitations in real-world settings.
• ➔ Proved formal statistical guarantees (e.g., bounds on ARL and detection delay) using tools like martingale analysis and convex optimization, bridging theoretical rigor with ML practicality.

Oct 2024 – May 2025 · 7 mos · Ann Arbor, Michigan, United States

• Project: LLM Watermarking using Sequential Detection
• ➔ Proposed and implemented a sequential Monte Carlo testing framework for detecting statistical watermarks in LM outputs, with anytime-valid guarantees and reduced computational cost.
• ➔ Replaced traditional permutation-based tests with a "testing by betting" strategy, resulting in significant reductions in the permutations required to make decisions while controlling Type-I error.
• ➔ Developed a modular pipeline comprising watermark generation, detection, and sequential hypothesis testing, and evaluated performance using OPT-1.3B on the C4 dataset.
• ➔ Demonstrated robustness of the sequential test under adversarial token-level perturbations with empirical power > 95% and controlled false-positive rate (α ≈ 0.05).

Aug 2024 – Apr 2025 · 8 mos · Ann Arbor, Michigan, United States

• F24: EECS-455: Wireless Communication Systems
• W25: EECS-330: Introduction to Antennas and Wireless Systems

Jan 2024 – Jul 2024 · 6 mos · Hyderabad, Telangana, India · On-site

• Team: Smart Manufacturing & Artificial Intelligence
• ➔ Suggested a new metric to analyze data & develop a troubleshooting system for fab machinery, reporting ~0.5% enhancement in yield.
• ➔ Fine-tuned a Llama model to produce an application automating frontend widgets in Angular, Highcharts, and AG Grid, resulting in an efficiency boost; contributed to confluence documentation.
• ➔ Implemented and optimized ETL pipelines, integrating APIs that enabled seamless data extraction from Snowflake to Micron's internal analytics tools, resulting in a 40% reduction in data processing time.

May 2023 – Jul 2023 · 2 mos · Edmonton, Alberta, Canada · On-site

• PI and Lab: Prof. Ashwin Iyer at M3 Lab
• Project: Developing Metasurface Liners for Improving Magnetic Resonance Imaging
• ➔ Incorporated meta materials to raise the cutoff frequency by a notable 15%, resulting in improved MRI clarity and signal propagation.
• ➔ Determined transmit efficiency & SNR and SAR profiles at 3.8 and 5.2 GHz in simulation using full-wave simulators & experiments.
• ➔ Developed a high-precision predictive model using XGBoost to determine the geometry of patches, achieving a 95% accuracy in solving optimization for EM characteristics and reducing decision making time.

Dec 2022 – Mar 2023 · 3 mos · Kolkata, West Bengal, India · Hybrid

• Project: Development of Multilayer Intelligent Reflecting Surfaces (IRS) for 6G Communications
• ➔ Achieved beam steering & EM properties modification through voltage variable capacitor diodes integrated into the 32 x 32 IRS unit.
• ➔ Trained an advanced ML model using Support Vector Machines (SVM) that achieved a 95% accuracy in predicting unit cell geometry.
• ➔ Innovated and designed IRS meta surface multi layer unit cells incorporating diodes for Ka-band applications, leading to a 20% increase in bandwidth and optimizing signal modulation.

Jun 2022 – Sep 2022 · 3 mos · Ahmedabad, Gujarat, India · Hybrid

• Project: Radar Cross Section Reduction using electronically employable methods
• ➔ Measured different types of space bound antennas in the Anechoic chambers at ATMD-ISRO
• ➔ Implemented a hybrid DL model (CNN-LSTM) for signal processing & performed MATLAB simulations on time-domain and frequency-domain signals, enhancing data analysis capabilities, proving the jet to be theoretically 0 RCS.
• ➔ Proposed integration of a Phased Array for beam steering with Mechanical Steering, resulting in a 50% increase in Field of View (FoV).