Research
NSF-SAFE OSE Project
Aug 2025 - Present
Built an automated vulnerability scanning pipeline for Jetstream2, scanning 118+ public VM images with Trivy and storing results in a centralized database for cross-image analysis. Extended it to map vulnerabilities to running research workloads and detect non-CVE dependency risks using Semgrep and ecosystem APIs.
Assessing the Vulnerabilities in Large Language Model (LLM)-generated Code: A Benchmarking Study
Aug 2025 - Apr 2026
Published at IEEE SAGAI '26, this study introduces LLMVulnBench and evaluates 2,400 agentic code-generation scenarios across four languages, application domains, and prompt strategies. Static analysis identified 1,115 deduplicated findings across 32 CWE categories, while structured prompts reduced vulnerability prevalence by 61% compared with naive prompts.
Using Transformers and DL with Stance Detection to Forecast Crypto Price Movement
May 2022 - Dec 2022
Developed a stance-detection-based forecasting system that predicted Bitcoin price movements by combining transformer-based NLP and deep learning time-series models. Achieved 80% stance-detection accuracy and a mean absolute error of $1,144 on price prediction.