Graduate student at Virginia Tech researching AI Reliability and Alignment: how LLMs and AI agents reason, and how they fail. Before grad school I spent ~7 years as a software developer at Fiserv, building merchant products and large-scale cloud systems.
🌐 Portfolio · 💼 LinkedIn · ✍️ Medium · 📄 Resume
- TwoWorlds: a benchmark for understanding how AI agents fail on scientific reasoning tasks
- Underspecified math reasoning: training methods for LLMs working on incomplete optimization problems
| Project | What it is |
|---|---|
| Training Qwen with GRPO | Team project: reinforcement learning for medical reasoning with Qwen2.5-3B (course report) |
| llm-from-scratch | Building large language models from first principles |
| Deep-learning | Vision, sequence models and CNN benchmarking experiments · site |
| Loblaw_Bio_Drug_Development | Clinical-trial immune-cell analysis with statistical testing and an interactive dashboard |
| asphalt_reimagined | Playing a racing game with hand gestures · demo |
ML: PyTorch · Hugging Face · RL for LLMs (GRPO) · Python
Engineering: Java · Spring Boot / Spring Cloud · microservices · React · cloud systems

