Undergraduate CS Researcher & Systems Builder

Moshfiqur Rahman Nayeem

Computer Science & Technology @ China University of Petroleum (Beijing)

About

I'm an undergraduate studying Computer Science and Technology at China University of Petroleum (Beijing) under the CUPB First-Class Scholarship for International Students.

My work centers on systems engineering, terminal productivity tooling, and multimodal medical AI. I prefer build-first learning, minimal architectures, and high-leverage tools that automate developer workflows from the Linux shell to resource-constrained deep learning inference.

EDUCATION
B.Eng. Computer Science ('27)
FOCUS
Multimodal AI & Systems
ENVIRONMENT
Arch Linux · Neovim · Zsh

Featured Work & Research

brain.zsh Open Source · Shell

Context-aware terminal layer for Zsh. Integrates state machines, project detection, automated LLM routing, and structured compiler error parsing — lazy-loaded in ~30ms.

  • Zero-re-scan cache: Persistent walk-up project detection across Rust, Python, Go, Node, and Docker environments.
  • Intelligent AI context routing: Compiles git state, branch history, and execution context for CLI AI tools (opencode, claude, llm).
  • Structured error parsing: Regex engine for Rust, Python, JS/TS, Go, and shell stack traces, offering actionable fix suggestions.
Zsh CLI Tooling LLM Routing Lazy Loading MIT
Hierarchical Lead-Aware Multimodal ECG-RAG
Thesis Research · In Progress

Explainable Myocardial Infarction diagnosis on 12-lead Electrocardiograms (School of Artificial Intelligence, CUPB). Status: Proposal approved; baseline experiments on PTB-XL in progress.

  • Hierarchical Lead-Instance Pooling: Aims to address spatial lead-blindness in 12-lead ECGs by preserving anatomical territories (inferior, anterior, lateral).
  • Guideline-Grounded Retrieval: Explores RAG pipelines grounding model predictions in ESC/AHA acute coronary syndrome guidelines to reduce clinical hallucinations.
  • Resource-Constrained Optimization: Designed for lightweight deployment and reproducible evaluation on standard PTB-XL benchmarks.
PyTorch Multimodal AI Medical RAG PTB-XL Research

Engineering Philosophy

I follow the Ponytail Principle (the practice of choosing the simplest, most minimal solution that genuinely works): the cleanest code is often the code you didn't have to write. Prioritize standard libraries, native OS features, and minimal parameter additions over speculative abstractions and architectural bloat.