CV

Download PDF

Contact

• Email: zixinwan04@gmail.com
• GitHub: stanzixinwan
• LinkedIn: zixinwan
• Website: stanzixinwan.github.io

Education

• Columbia University, New York, NY
  M.S. in Computer Science, Aug 2026 – Dec 2027 (expected)

• Brandeis University, Waltham, MA
  B.S. in Computer Science and Biology, Aug 2022 – May 2026
  GPA: 3.76 / 4.00, magna cum laude, Dean’s List
  Coursework: Machine Learning, Natural Language Processing, Linear Algebra, Statistics, Data Structures & Algorithms, Computer Systems, Operating Systems, Theory of Computation

Research

• Honors thesis, Van Hooser Lab (advisor: Stephen D. Van Hooser): functional decoding and neural tuning dynamics in visual cortex. Manuscript in preparation. Research
• Visiting student researcher, Shanghai Jiao Tong University (advisor: Yutong Ban): multimodal VLA training and dual-arm data infrastructure, including +18% cross-task success on InternVL-1B (8×A100).

Experience

• Visiting Student Researcher, Shanghai Jiao Tong University, Shanghai, China
  June 2026 – Sept 2026
  • Engineered a high-throughput multimodal data pipeline synchronizing dual-arm joint states and 3-channel RGB camera feeds at 30 FPS, reducing I/O latency by 35% across 100+ manipulation episodes.
  • Scaled SFT and advantage-conditioned policy training for an InternVL-1B vision-language-action model.
  • Improved cross-task policy success rates by 18%, using PyTorch FSDP on an 8×A100 GPU cluster.
  • Built a low-latency human-in-the-loop teleoperation and intervention framework for closed-loop evaluation.
• Machine Learning Research Assistant, Van Hooser Lab, Brandeis University, Waltham, MA
  Sept 2024 – May 2026
  • Authored and defended an honors thesis on functional decoding and neural tuning dynamics in visual cortex.
  • Architected an automated scientific compute pipeline in Python/MATLAB for 2-photon calcium imaging data, transforming terabytes of raw fluorescence streams into memory-mapped binary representations.
  • Developed 50+ modular feature extraction scripts, cutting data processing time by 40%.
  • Maintained CI/CD workflows for open-source scientific tools (vhlab-toolbox and NDI-matlab).
• Software Engineer Intern, Noah AI, Remote
  May 2024 – Aug 2024
  • Designed and deployed an enterprise hybrid RAG pipeline combining dense vector embeddings (FAISS) with BM25 sparse keyword search over medical literature, improving retrieval and reducing hallucinations by 15%.
  • Implemented Cross-Encoder re-ranking and structured JSON Schema validation layers, enforcing output format constraints across high-concurrency LLM agent workflows.
  • Developed backend microservices using FastAPI and Next.js modules, optimizing API response times.

Projects

• DistIE: Distributed Inference Engine · Go, C++, Python, gRPC, pybind11
  Building a multi-node LLM inference stack: a Go gateway for concurrent streaming requests, a C++ core for block-based KV-cache and GPU memory pooling, and Python orchestration for continuous batching. GitHub

• Cross-Domain Text-to-SQL System · PyTorch, Transformers, LoRA/QLoRA, FAISS, Docker, FastAPI
  Fine-tuned encoder-decoder models with QLoRA on the Spider benchmark, reaching 82% execution accuracy on unseen multi-table schemas. Built a schema-aware RAG pipeline with FAISS and containerized inference behind asynchronous FastAPI endpoints.

• LLM-Powered Nutrition Tracking Platform · Python, TypeScript, PostgreSQL, Prisma, Vitest
  Built an AI agent orchestration system that parses unstructured dietary text into validated macronutrient structures. Designed a high-concurrency PostgreSQL schema with Prisma and CI/CD with Vitest and GitHub Actions (90%+ coverage).

Skills

• Programming: Python, Java, C/C++/C#, JavaScript, SQL, Go, HTML/CSS
• Frameworks: React, Next.js, FastAPI, Node.js, Prisma ORM
• Dev & infra: Git, GitHub Actions, Linux, CI/CD, Vitest, PostgreSQL, Docker
• Libraries: PyTorch, FAISS, scikit-learn, Pandas, NumPy