Projects
A deeper dive into the security, ML, and systems work I have shipped recently.
Chrome Extension · Python, Flask, React, Azure OpenAI · Dec 2024 – Feb 2025
- Ships an in-browser AI chatbot that summarizes visited pages, answers questions, and quotes sources.
- Backed by a Flask inference API that orchestrates Azure OpenAI models, embeddings, and RAG flows.
CS180 Project 4 · PyTorch, NeRF, Rendering · 2025
- Implemented positional encodings, hierarchical sampling, and volumetric rendering per the UC Berkeley CS180 brief.
- Reconstructed novel 3D views from sparse multi-view captures, tuning with gradient clipping and LR warmups.
- Authored a comprehensive report comparing against baseline NeRF outputs and documenting experiments.
e2e-cyber-bench
Security Agent Benchmark · 1,507 Tasks · 2025
- Curated a suite of historical CVEs linked to reproducible build/test scripts for automated evaluation.
- Designed scoring harnesses for tasks spanning static analysis, exploit development, and patch validation.
- Adopted across our research groups to compare LLM-based fuzzing and remediation agents.
LLM-Driven Fuzzing Seeds · 2025 · Published at LLM4Sec Workshop
- Automates seed-corpus creation with LLM reasoning to increase fuzzing across OSS-Fuzz projects.
- Containerized OSS-Fuzz, LibFuzzer, and Magma benchmark pipelines for multi-corpus experimentation.
- Open-sourced with configuration recipes so other labs can reproduce the workshop results.
CodaBot
Secure File Sharing · Go, AES-CTR, RSA · Team of 2 · Sep 2024 – Nov 2024
- Designed a threat-model aware storage layer with HMAC, hashing, and RSA key wrapping for user sharing.
- Engineered constant-bandwidth append operations that maintain confidentiality even under active attackers.
Deep RL for Green Crab Mitigation · RecurrentPPO, Gymnasium · 2024
- Extended the open-source codebase with custom Gymnasium environments, Optuna tuning scripts, and dashboards.
- Demonstrated sustainable harvesting policies using RecurrentPPO, including robustness checks with noisy observations.
Flow Matching U-Net · MNIST · 2024
- Implemented a diffusion-style flow-matching model with time and class conditioning in a custom U-Net.
- Trained on MNIST digits to generate class-controlled samples from pure noise.
- Evaluated generation quality with FID and ablations on conditioning strategies.
Post
Reddit-like Social Platform · React, Node.js, Firebase · Team of 4 · Dec 2023
- Implemented authentication, post creation, and sharing feeds with Firebase Auth/Firestore.
- Led front-end architecture, responsive styling, and CI deployment for classroom showcase.
Build Your Own World (BYOW)
2D Maze Exploration Game · Java · Team of 2 · Apr 2023 – May 2023
- Developed procedurally-generated tile worlds using DFS-based maze construction and OOP design patterns.
- Shipped avatar controls, theming, and save/load functionality for repeatable gameplay sessions.