Summary
Overview
Work History
Education
Timeline
Research Experience
Current Research (Intended for Publication)
Awards and Recognition
Research & Class Projects
Generic

Yevin Goonatilake

Leesburg,VA

Summary

PhD Computer Science researcher at George Mason University (Presidential Scholar) specializing in LLM security, watermarking, and cryptography. Extensive experience building end-to-end agentic AI systems and using AI tools to rapidly design, prototype, and ship research and software systems. Strong background in LLM/ML/DL development, including fine-tuning, prompt engineering, and technical evaluation. Currently co-authoring a few publications in preparation targeting top-tier conferences (USENIX, Euro S&P, NeuroIPS).

Overview

3
3
years of professional experience

Work History

LYT Systems, Inc.

Software Engineer Intern
Leesburg, VA
10.2024 - 02.2026
  • Developing and integrating a full-stack resume submission feature for the company website
  • Built responsive front-end forms and UI components using Angular
  • Implemented back-end RESTful APIs using Java Spring Boot to process incoming applicant data, perform validation, and interact with the database
  • Designed and utilized a MySQL database schema to efficiently store and manage resume details

GMU Computer Science Department

Computer Science Undergraduate Teaching Assistant
Fairfax, VA
01.2024 - 12.2024

Synthesis of Ethics and Law for Computer Professionals & Computer Systems and Programming

  • Graded student's semester long projects
  • Lead teaching assistant
  • Created numerous content to aid and guide future teaching assistants
  • Conducted a extensive legal research project, generating optimal case files and procedures
  • Debugged and analyzed student code, providing help and suggestions

GMU Mathematics Tutoring Center

Mathematics Tutor
Fairfax, VA
01.2023 - 05.2024
  • Assisted students in completing coursework through virtual sessions.
  • Addressed learning hurdles through personalized, one-on-one lessons
  • Administered and managed virtual Math tutoring services at GMU, as the main online tutor

Education

Ph.D. - Computer Science

George Mason University
Fairfax, VA
12-2028

Master of Science - Computer Science

George Mason University (GMU)
Fairfax, VA
05-2025

Bachelor of Science - Computer Science

George Mason University (GMU)
Fairfax, VA
12-2024

Timeline

LYT Systems, Inc.

Software Engineer Intern
10.2024 - 02.2026

GMU Computer Science Department

Computer Science Undergraduate Teaching Assistant
01.2024 - 12.2024

GMU Mathematics Tutoring Center

Mathematics Tutor
01.2023 - 05.2024

Ph.D. - Computer Science

George Mason University

Master of Science - Computer Science

George Mason University (GMU)

Bachelor of Science - Computer Science

George Mason University (GMU)

Research Experience

The Coding Limits of Robust Watermarking for Generative Models, Spring 2025 - Fall 2025 Semesters

Advised by Dr. Giuseppe Ateniese (GMU) and Dr. Daniele Venturi (Sapienza University of Rome)

  • Built a coding-theoretic model for robustness and tamper detection in generative-model watermarking
  • Proved sharp upper bounds showing robust watermark recovery fails past 1-1/q corruption (q-ary)
  • Independently discovered a revolutionary simple real-world break: crop-and-resize disrupts latent watermark structure and defeats decoding while preserving image quality, rendering the watermark scheme we test on useless
  • Linked theory to practice by showing the limit appears in all standard watermarking pipelines

Energy Demand Prediction & Optimization - GMU Power Consumption

Decision Guidance System Research, Spring 2025 - Summer 2025 Semesters

  • Developed ML models to predict GMU peak electricity demand from consumption history, weather, and university event data
  • Engineered features and performed dataset integration/QA to improve signal quality and model stability
  • Benchmarked candidate models and tuned hyperparameters to finalize a high-performing predictor
  • Delivered forecasts to inform purchasing decisions aimed at reducing operational energy costs
  • This research was for reducing university operational costs by using efficient energy estimation based on our model's demand predictions

Current Research (Intended for Publication)

Detecting “Undetectable” Watermark Removal 

  • Identified a practical gap in watermark removal: removal pipelines often leave learnable forensic artifacts, even when the watermark is destroyed
  • Trained a detector that recognizes watermark-removed images across multiple pipelines and schemes
  • Conducted an evaluation in order to confirm detection regardless of concealment attempts (re-encodes, tampering, normalization, and ablations)
  • Demonstrated generalization regardless of scheme or scenario; extending to a formal study of undetectability limits for removers

Cross-Prompt Logit-Based Language Model Reasoning Distillation

  • Co-developed a distillation approach that transfers reasoning behavior from a larger, superior model to a smaller LLM
  • Engineered, optimal prompts improved accuracy on MATH dataset by ~18%; showed these gains can be distilled rather than prompt-only
  • Top-K logit distillation produced a large jump over SFT (mid-60s to low-80s percent accuracy), one of the current renowned techniques
  • Used an external LLM judge and strong baseline comparisons to assess the method’s impact and measure accuracy improvements

Attacking “Unlearnable Audio” Protections (MusicShield)

  • Conducting a security evaluation to break the MusicShield (IEEE S&P ’25) scheme: imperceptible perturbations to make music “unlearnable”
  • Evaluating robustness across model families and attacker capabilities, emphasizing transfer and real-world plausibility
  • Implementing both preprocessing-based purification and learned denoisers to remove protection while preserving audio quality
  • Developing adaptive fine-tuning attacks to recover learnability from protected samples.

Awards and Recognition

  • Became the youngest to ever receive a Master of Science in Computer Science, from any university (at age 16)
  • Youngest Bachelor of Science to graduate from George Mason University (at age 15)
  • Youngest to graduate with an Associate's degree from Northern Virginia Community College (at age 13)
  • One of only four Presidential Scholarship recipients in the current four year cycle at GMU
  • Distinguished Academic Achievement recipient for both Bachelor of Science and Master of Science in Computer Science
  • The youngest player in the most competitive adult nationwide cricket tournament, Minor League Cricket, representing Baltimore Royals

Research & Class Projects

  • Development & Research Analysis of AI Financial Quant Agent (Agentic Workflow System)
  • Robustness Analysis of Text Watermarking for Large Language Models
  • Harvard OS/161 Operating Systems Projects (Syscalls, Concurrency, and Systems Programming)
  • Unix Shell + Job Control Implementation (Pipelines, Redirection, Task Management)
  • Software Web Design: Full-Stack Interactive Campus Map (Search + Google Maps API)
Yevin Goonatilake