GPU Systems Research · Hardware Verification · Agentic AI Infrastructure
CS @ University of Central Florida · Unary Lab (Prof. Di Wu) · Graduating Spring 2026
I study how AI systems use hardware — from GPU thermal behavior to agentic workload profiling to autonomous hardware verification.
| Project | What it does | Stack |
|---|---|---|
| Project Ava | Autonomous AI agent for Verilog verification. 19 designs · 191 tests · 100% pass rate · Mutation testing · SHA-256 ASIC-proven | Python, Verilog, cocotb, Icarus Verilog |
| AgentTrace | Agentic workload profiler. 43.3% GPU idle — validates Kim et al. (HPCA 2026) | Python, ECharts, Supabase |
| ThinkTank | Crowdsourced reasoning labels + PRM. Qwen2.5-0.5B + LoRA — 95.7% accuracy | PyTorch, TRL, PEFT, HuggingFace |
| ThermalTrace | GPU thermal monitoring. 63,700+ samples. 25°C spread on 4× RTX 4070 Ti | Python, pynvml |
- AMD MI300X profiler trace analysis via Chopper — DVFS-driven straggler coupling
- GPU thermal characterization for distributed training
- Agentic AI workload profiling — CPU/GPU utilization, token economics
Children's EdTech streaming platform · 6 production repositories
- Recommendation engine + audio library across iOS/Android/Web
- BigQuery pipelines, Looker Studio dashboards, GTM tracking, RevenueCat webhooks
- Performance: N+1 query fixes, Redis caching, PostgreSQL indexing
- 1,100+ commits · 265 PRs · 103 Jira tickets
| Type | Link |
|---|---|
| 🤖 Model | thinktank-prm-qwen2.5-0.5b |
| 📊 Dataset | thinktank-reasoning-labels |
| 📊 Dataset | thinktank-reward-pairs |