Full-stack · Machine learning · Computer vision
Moeur Award (top 1%) · IEEE Scholar
I build systems that are useful without being noisy — production APIs, ML pipelines, and computer vision tools with a focus on speed, clarity, and interfaces calm enough to trust.
Recent work spans a co-op building Redis-backed API platforms at Bytewerx, research on LLM-powered robotic navigation at ASU's LLEAS lab, and independent projects across NLP, computer vision, and full-stack deployment.
Open to roles in full-stack engineering, ML systems, and applied AI.
| project | what it does | stack | |
|---|---|---|---|
| 01 | CorpusForge | 54M-parameter GPT decoder built from scratch — trained on 50M tokens of OpenWebText with no pretrained weights. Flash Attention, cosine LR with warmup, AdamW, served via FastAPI | PyTorch Flash Attention FastAPI |
| 02 | Onsight | End-to-end NLP system over 1,000+ scraped wiki pages — zero-shot theme classification, NER character network extraction, fine-tuned HuggingFace classifier, all surfaced in a Gradio dashboard | HuggingFace SpaCy Scrapy Gradio |
| 03 | Flock | Full-stack ChatGPT-style chatbot with live OpenAI API integration, persistent session handling, and a production-grade UI — deployed on AWS | React Node.js TypeScript AWS |
| 04 | SoccerSense | Computer vision pipeline achieving 82% accuracy on real-time player and team stat generation — YOLOv8 tracking, KMeans segmentation, perspective transforms | YOLOv8 OpenCV PyTorch |
| 05 | SyncLink | Distributed NFS in C supporting 50 concurrent clients across a three-tier architecture — binary 9-byte protocol, full CRUD, cross-server copy | C POSIX Sockets |
Python TypeScript C C++ Java React Node.js Angular Django Flask PyTorch OpenCV HuggingFace Scrapy AWS Docker Redis MongoDB PostgreSQL NumPy Pandas
桜 · calm systems · 桜