Professional Summary
I am a Ph.D. candidate at Florida International University, working as a Research Assistant at the Systems Security Lab (SecLab) under the supervision of Dr. Amin Kharraz. My research focuses on web and browser security, with an emphasis on applying data-driven and machine learning techniques to detect emerging threats at scale, analyzing LLM-based browser agents and automated interactions in the age of the agentic web. I also bring over five years of industry experience as a Site Reliability Engineer and Software Engineer, providing a strong foundation in building secure and scalable systems.
Education
Ph.D. Computer Science
Florida International University
M.Sc. Computer Engineering
Sharif University of Technology
B.Sc. Computer Engineering
University of Tehran
Professional Experience
- Developed LLM-powered Browser Agents to bypass complex web challenges such as reCaptcha with over 90% success rate, demonstrating critical detection requirements in modern agentic web.
- Implemented ML pipelines to classify users from browser interaction data, creating Time Series data collection, feature extraction, and model training using PyTorch, Scikit-Learn, and LLM services.
- Built a text Captcha solving framework with 80% success rate in real-world websites in a single attempt by using Tensorflow Object Detection trained on data from over 1M websites.
- Analyzed ecosystem-wide Vulnerability Trends by unifying GitHub Advisory and Snyk.io reports across 10+ package managers, uncovering CWE patterns, ecosystem differences, and malicious packages.
- Detected zero-day anomalies within encrypted network flows by implementing Open Set Recognition models in TensorFlow, leveraging feature extraction on terabytes of PCAP data.
- Led research projects in collaboration with faculty, students, and external teams, including Microsoft Security, contributing to open-source tools for user behavior analysis and phishing detection.
- Ensured 99% uptime for 60 microservices by maintaining the infrastructure of over 100 Ubuntu servers for compute clusters and databases across Kubernetes, PostgreSQL, and MongoDB.
- Led platform team to design and build Node.js and Python gRPC and Socket message passing libraries with integrated Prometheus and healthcheck probes for full observability in Grafana dashboards.
- Accelerated rapid adoption of new technologies such as MinIO, Vault, Kafka, and Spark, researching and deploying production-ready solutions to enhance platform reliability and data processing capabilities.
- Reduced deployment errors and established automated rollouts by implementing CI/CD pipelines to include canary-testing, staging, and production environments, utilizing Bash and Ansible.
- Set up and maintained runtime environment and deployment pipeline using Java app engine on Google Cloud Platform.
- Implemented data storage infrastructure including PostgreSQL and Kafka for scalable data management.
- Developed a Deep Learning video analysis pipeline (OpenCV and CNN) to process 100+ hours of news broadcast data.
- Enabled Multimodal Modeling of temporal visual data for studio scene frame classification.
- Implemented automated Face Detection pipeline with 90%+ precision for frame-level annotation.
- Developed backend using Express.js to provide REST APIs for users, payments, and request matching.
- Designed and developed frontend using jQuery for request submission forms.
- Implemented Telegram Bot for automated notifications to agents.
Projects
Analyzing LLM-Based Browser Agents in Captcha Solving
Evaluated popular Captcha defenses against modern LLM agents, exposing limitations in trust mechanisms and detection. Built an automated solver using LLM-based browser agents to simulate human behavior and successfully bypass reCaptcha v3 in real-world deployments.
LLM-Powered Clinical Trial Matching
Developed an LLM-driven pipeline for eligibility matching with clinical trial protocols, leveraging Prompt Chaining to process 30K+ trial documents. Implemented Chat-with-PDF feature using RAG for interactive querying.
Analyzing Adversarial Payloads via Large Language Models
Collected and analyzed a corpus of 90K+ payloads from publicly exposed online forms. Performed comprehensive analysis using Open-source Large Language Models for payload classification.
Investigating Security Challenges in Open-Source Software
Analyzed vulnerability trends across 10+ ecosystems using GitHub Advisory and Snyk.io data. Identified common vulnerability types and compared distribution patterns, with detailed analysis of malicious packages in the NPM ecosystem.
Real-Time Browser Interaction Modeling
Developed a scalable pipeline for Multi-Modal browser interaction data collection and classification of 1.6B+ artifacts. Published research demonstrating advanced bot detection capabilities.
View PublicationEvaluating Resilience of Behavioral Bot Detection
Implemented Adversarial Machine Learning attacks (FGSM, Genetic Algorithm) to evaluate deep learning model robustness in practical scenarios.
View PublicationAdversarial Assessment of Text Captchas
Developed Object Detection module using Tensorflow to solve real-world text-based Captchas, achieving 80%+ success rate on samples from 1M+ top websites.
View PublicationPhishing Websites Detection
Investigated code reuse patterns in 300K+ phishing websites using Semi-Supervised Clustering, enabling scalable detection of malicious websites with similar origins.
Deep Open Set Analysis for Network Traffic
Built Feature Engineering pipelines for terabytes of PCAP data and implemented Open Set Recognition models in TensorFlow to detect zero-day anomalies in encrypted flows.
View PublicationPublications
Cracking the Web: Analyzing LLM-Based Browsers and Automated Solvers in the Web
WebGuard: Detecting Evasive Web Scanners via a Multi‑Modal Forensics Engine
Open Source, Open Threats? Investigating Security Challenges in Open-Source Software
EnSolver: Uncertainty-Aware Ensemble Captcha Solvers with Theoretical Guarantees
Breaking the Bot Barrier: Evaluating Adversarial AI Against Multi-Modal Defenses
The Matter of Captchas: An Analysis of a Brittle Security Feature on the Modern Web
An End‑to‑End Analysis of Covid‑Themed Scams in the Wild
An Adaptable Deep Learning‑based Intrusion Detection System to Zero‑day Attacks
Honors and Services
Conference Reviewer
RAID, DIMVA (20-25% Acceptance Rate)
2024, 2025
Journal Reviewer
IEEE Transactions on Information Forensics & Security (TIFS, IF 8.0)
2024
Scholarship Recipient
Student Government Association Graduate Scholarship, FIU
2024
Teaching Assistant
Software Security, Operating Systems, and Python Programming, FIU
2021-2025
Mentor
Research methodologies, coding practices, and project planning for graduate students
2022-2025
Cloud Expertise
Amazon Web Services
