CV

Graduate Research Assistant

AI and Computing Lab

Conduct research in Explainable AI (XAI), developing attribution methods and evaluation metrics for deep neural networks with applications in adversarial robustness and out-of-distribution detection for high-stakes deployments.

• Developed Metric-Driven Attributions (MDA) framework outperforming 7 ViT attribution methods by 12% across 12 metrics on ImageNet, establishing state-of-the-art in transformer interpretability (ICLR 2025).
• Developed attribution graphs to enable an understanding of how causal information propagates from an LLM prompt to its generation, improving faithfulness on SOTA LLMs by 20% on average (Preprint 2025).
• Pioneered Magnitude Aligned Scoring (MAS) XAI evaluation benchmark 4x more sensitive to attribution changes, 2x more consistent, 1.6x more baseline-invariant than existing metrics (IJCAI 2024).
• Designed Integrated Decision Gradients (IDG), a novel path-based attribution to address gradient saturation in deep networks, achieving state-of-the-art performance (AAAI 2024).
• Created InFlow, an information flow-based explanation framework for ViTs outperforming 6 attribution methods by up to 18% in eight metrics (AISTATS 2025).
• Developed attribution-based adversarial detection system achieving 99% accuracy classifying pixel and patch attacks in CNNs trained on ImageNet; supported by DARPA (MILCOM 2023).
• Invited talk: “Towards Effective XAI of Modern Models”. Prince of Songkla University, Thailand (2024).
• Best Poster Award: 2025 Nelms IoT Conference.
• Mentored 12 undergraduate researchers; outcomes include 3 honors theses and 2 graduate admissions.
• Reviewed for ICLR, ICML, IJCAI, CVPR, ECCV, and JMLR.
• Volunteered for the 2024 and 2025 Nelms IoT Conferences.

Back-End Developer Intern

Avanade/Accenture

• Contributed to 10-person development team building enterprise CRM system for field service management.
• Architected and deployed CI/CD pipeline using Azure DevOps, reducing deployment time and increasing release reliability for client-facing applications.
• Automated business workflows using Power Automate and Azure services, eliminating weekly manual processing and improving customer response time.

Full Stack Developer & Electrical Engineering Intern

Limbitless Solutions

• Built website from design mockups as sole developer (Bootstrap/HTML/CSS/JS), serving as primary platform for donor outreach and patient enrollment; 80\% of code remains in production after 7+ years.
• Built internal project and people management tools as sole developer, streamlining workflow coordination for 30+ person engineering team.
• Designed custom HCI PCBs enabling gamified learning for prosthesis control.
• Conducted human-subjects studies validating user experience improvements for pediatric patients.

Doctor of Philosophy

University of Florida

Computer Engineering

• Expected Graduation in August of 2026
• Dean’s Research Award

Master's of Science

University of Central Florida

Computer Engineering

• Graduate Dean’s Fellowship
• Graduate Presentation Fellowship
• Alireza Seyedi Innovation Endowed Scholarship

Bachelor's of Science

University of Central Florida

Computer Engineering

• Burnett Honors College Graduate

Metric-Driven Attributions for Vision Transformers

ICLR

Explaining ViTs Using Information Flow

AISTATS

Detecting and Removing Adversarial Patches using Frequency Signatures

IJCNN

Out-of-Distribution Detection for Contrastive Models using Angular Distance Measures

ICMLA

Attribution Quality Metrics with Magnitude Alignment

IJCAI

Integrated Decision Gradients: Compute Your Attributions Where the Model Makes Its Decision

AAAI

Adversarial Pixel and Patch Detection Using Attribution Analysis

MILCOM

Explaining the Reasoning of Large Language Models Using Attribution Graphs

Conference Paper Under Review (Preprint Available)

Verifiable Natural Language to Linear Temporal Logic Translation: A Benchmark Dataset and Evaluation Suite

Conference Paper Under Review (Preprint Available)

Grounding Natural Language into Formal System Signatures for Temporal Logical Translation

Conference Paper Under Review (Preprint Available)

Question Decomposition using Masked Language Modeling for Knowledge Editing

Conference Paper Under Review

Are VLMs 3D-Aware? A Study of Scene Generation via Structured Tokenization

Conference Paper Under Review

Explainable AI in Image Classification: A Comprehensive Survey of Attribution Methods and Metrics

Journal Paper Under Review

Rethinking Attribution Quality Metrics for Vision Transformers

Journal Paper Under Review