Graduate student at San José State University working on AI security, LLM applications, and deep learning.
- 🏆 6th prize, AI Security Engineering Hackathon (Snyk × AWS × Guild.ai) with AgentBreaker
- 🔐 Interested in how LLM agents break: prompt injection, tool poisoning, excessive agency
- 🧠 Building with LLM tool-calling, Transformers, and EEG/brain-computer interfaces
| Project | What it is | Stack |
|---|---|---|
| 🛡️ AgentBreaker | Gandalf-style game where you hack 5 deliberately vulnerable AI agents to learn prompt injection, MCP tool poisoning, excessive permissions, data leakage, and command injection. Snyk score 10/10. | Python · FastAPI · Guild.ai · Snyk |
| 🎓 AI-Assistant | Tool-calling course-planning assistant grounded in a SQLite database. Compares Phi-3.5 vs Llama-3-8B (86% tool-selection accuracy) using meta prompting, chaining, self-reflection, and caching (~5x speedup). | Python · Hugging Face · Gradio · SQLite |
| 🧠 BrainWave | Hybrid CNN-Transformer that classifies imagined movements from 64-channel EEG: 72.3% accuracy across 103 subjects with 4.2 ms real-time inference. | PyTorch · MNE · scikit-learn |
| ♻️ E-waste Monitor | Tracks device lifespans, raises replacement alerts, and suggests responsible recycling routes. | Python |