Steffi Roy
Actively seeking PhD positions · Fall 2026

Steffi Roy

MS student at Northeastern interested in machine learning systems, safe AI, and efficient deep learning. Previously: hardware security research at UF, silicon verification at Texas Instruments and Microsoft.

ML Systems & Efficient Deep Learning

Model compression, parameter-efficient fine-tuning, and hardware–software co-design for scalable machine learning workloads.

Safe & Certifiable AI

Robustness under distribution shift, out-of-distribution detection, and formal guarantees for learned systems — particularly in autonomous and safety-critical settings.

Hardware Security for AI

Side-channel resilience for neural network inference, privacy-preserving computation, and secure hardware architectures for deployed ML systems.

Garbled EDA: Privacy-Preserving Electronic Design Automation

M. Hashemi, S. Roy, F. Ganji, D. Forte

ICCAD 2022

Designed a framework to run chip design tools securely without revealing proprietary IP, using cryptographic multiparty computation.

Active IC Metering Protocol Security with Oblivious Transfer

S. Roy, M. Hashemi, F. Ganji, D. Forte

SRC TECHCON 2022

Built and hardened protocols that prevent unauthorized cloning of chips, using oblivious transfer cryptography.

HWGN2: Side-Channel Protected Neural Networks

M. Hashemi, S. Roy, D. Forte, F. Ganji

SPACE 2022

Developed a hardware framework that lets neural networks run securely on FPGAs without leaking information through power/timing channels.

Multi-Agent Reinforcement Learning

MADDPG vs. DDPG  ·  Northeastern

Implemented and benchmarked MADDPG and DDPG in cooperative and competitive multi-agent environments. Analyzed policy convergence, reward shaping, and inter-agent communication under partial observability.

LLM Compression & Efficient Training

Northeastern

Explored parameter-efficient fine-tuning and model compression for large language models. Participated in the OpenAI parameter golf challenge — minimizing parameter count while preserving GPT training quality.

HabitBuddy

HCI Application  ·  Northeastern, CS 5340

Led end-to-end UX research and iterative design of a companion-based habit tracker — user personas, paper prototyping, heuristic evaluation, and high-fidelity Figma prototype.

Northeastern University — Khoury College of Computer Sciences

M.S. Computer Science  ·  2026 – Present

University of Florida

M.S. Electrical & Computer Engineering  ·  2021 – 2023

Mahindra Ecole Centrale, Hyderabad

B.Tech Electrical & Electronics Engineering  ·  2016 – 2020