Biography
Hi! I am a Computer Science student at UC Berkeley (Class of 2026) with a 4.0 GPA.
I focus on AI for security and work closely with Prof. David Wagner and
Prof. Dawn Song on autonomous agent benchmarking, large-scale fuzzing,
and real-time threat detection.
Outside of security, I working with Prof.Carl Bottiger in the Eric & Wendy Schmidt Center on reinforcement learning for environmental decision-making,
Research Experience
- Automated fuzz corpus collection with LLM to uncover vulnerabilities across OSS projects.
- Authored Scrapy + Search API tooling to locate seed files with diverse characteristics.
- Containerized OSS-Fuzz, LibFuzzer, and Magma benchmark pipelines for multi-corpus experimentation.
- Co-authored the SeedAIchemy paper for the 2025 LLM4Sec Workshop at ICDM.
- Generated LLM-based log rules that flag malicious host activity in real time.
- Lead developer for e2e-cyber-bench, a 1,500+ instance benchmark for evaluating AI agents on real-world vulnerability analysis tasks.
- Integrated historical CVEs from 188 large projects and automated reproducible setups for each target.
- Built fuzzing and unit-test harnesses that evaluate whether agents successfully patch vulnerabilities.
- Determined optimal green crab mitigation strategies with RecurrentPPO and other deep RL approaches.
- Fine-tuned agents and executed Optuna-based hyperparameter searches.
- Built Gymnasium simulations to visualize population dynamics over time.
- Crafted training environments with varied observations and randomness to improve robustness.
Working Experience
- Implemented a Python + MongoDB storage backend to accelerate metadata queries for the patch intelligence platform.
- Developed a PDF-to-text prototype with image and data extraction to improve downstream document ingestion accuracy.
- Completed a graphRAG workflow powered by Qwen, Microsoft graphRAG, and LM Studio for contextual retrieval.
- Benchmarked Text-to-SQL models on private datasets using Hugging Face and LlamaIndex.
- Designed a website for Floras showcasing 200 sustainable projects with searching and filtering features.
- Implemented the storage backend with Python and MongoDB for fast metadata queries from the Patch database.
- Developed front-end website integration using Flask, HTML, JavaScript, and CSS.
Publications
- Demonstrated how LLMs generate high-quality seed corpora that boost fuzzing code coverage and exploit discovery.
- Benchmarked across OSS-Fuzz targets and released evaluation tooling for the community.
Teaching Experience
- Ran weekly small-group sessions covering RISC-V, C, CPU design, and cache behavior.
- Prepared walkthroughs and practice problems for exam review sessions.
- Supported weekly teaching sections and helped 100+ students to reinforce core data-structure concepts.
- Triaged lab queues, debugged Java code, and curated study notes for upcoming exams.
Highlighted Projects
Chrome extension that pipes any page through Azure-hosted LLM + RAG pipelines, shipped with Flask APIs and a React UI.
A suite of CS180 projects: panorama building, NeRF reconstruction, and diffusion-model training, all documented in the linked portfolio.
1,500+ CVE-derived evaluation tasks with reproducible scripts for benchmarking autonomous security agents.
LLM-powered corpus generator that boosts fuzzing; a LLM4Sec workshop paper.
Secure Go-based file sharing system with HMAC, RSA, AES-CTR, and constant-bandwidth append algorithms.
DRL agents, Gymnasium environments, and Optuna tuning scripts for sustainable green crab invasion control.
detailed project list.
Skills
- Python (NumPy, pandas)
- Java
- C
- Go
- SQL / MySQL
- RISC-V
- HTML & CSS
- JavaScript
- React
- Docker
- Firebase
- MongoDB
- Hugging Face
- LlamaIndex
- LangChain
- PyTorch
- Scikit-learn
- Azure
- Linux