| Project | Description | |
|---|---|---|
| 📄 | From Refusal Tokens to Refusal Control | Categorical steering vectors for fine-grained LLM refusal control. −13.7% over-refusal, +14.2% harmful-prompt refusal on LLaMA-3 8B with zero capability degradation. 🎉 NeurIPS 2025 MechInterp Workshop · Under review ICML 2026 |
| 💠 | WaferDetect | Wafer-map defect-detection platform for semiconductor fabs — fine-tuned YOLO26x-seg segmenting 21 overlapping defect classes (0.852 mask mAP@50), with Poisson/Stapper yield modeling, per-defect cost estimation, and KLARF import/export |
| 🖊️ | Sketch2Graphviz | Converts hand-drawn graph sketches into executable Graphviz DOT code via LoRA fine-tuned Llama 3.2 11B Vision + RAG over a PostgreSQL/PGVector vector DB — 97.96% render success rate |
| ⚙️ | AutoNeuroNet | Reverse-mode automatic differentiation engine and neural network library built from scratch in C++ — published to PyPI with PyBind11 bindings and NumPy interoperability |
| 🔍 | GitLensAI | AI-powered Git intelligence tool for natural language querying of repository history, commit analysis, and developer insights |
| 📰 | TruthGuard | AI-powered misinformation and fake news detection with a fine-tuned DistilBERT classifier, FastAPI backend, React web app, and published Chrome extension |
- 📄 First author of a paper on LLM safety and refusal steering accepted to the NeurIPS 2025 Mechanistic Interpretability Workshop, currently under review at ICML 2026
- 🥇 Winner — 2023 Congressional App Challenge (Georgia's 5th District) with TruthGuard, an AI-powered fake news detector
- 🔬 ML Researcher at Emory University (Dr. Liang Zhao) — Latent Graph World Models · Research Contributor at CMU LTI's WAVLab (Dr. Shinji Watanabe)
- 🤝 Co-founder & Co-president of Science for Survival, a 501(c)(3) that has reached 1000+ students across the world.
ML / AI
Systems & Backend
Frontend
Open to research collaborations and impactful AI projects — reach out anytime.