DistIE: Distributed Inference Engine Permalink
Published:
Go gateway + C++ KV-cache/memory pool + Python continuous batching. A production-style LLM inference stack, not a notebook demo.
Portfolio
Systems and research first, then applied ML. DistIE and lab work are the pieces I would send a recruiter or a PI.Vision-Language-Action Policy Training
Published:
+18% cross-task VLA success on InternVL-1B; 35% lower I/O latency across 100+ dual-arm episodes (8×A100, SJTU).
Scientific Compute for 2-Photon Imaging
Published:
Terabyte-scale calcium-imaging pipelines and an honors thesis on neural tuning dynamics (Van Hooser Lab). Manuscript in preparation.
Cross-Domain Text-to-SQL
Published:
82% Spider execution accuracy with QLoRA fine-tuning and a FAISS schema-aware RAG serving path.
LLM-Powered Nutrition Tracker
Published:
Schema-validated LLM agents for dietary logs, PostgreSQL/Prisma, and 90%+ test coverage (Vitest, GitHub Actions).