AI Coding

I rely heavily on AI coding tools (Claude Code, Cursor, etc.) in my daily work and research. Here I document real-world cases and reflections.

Tools

• Claude Code — Primary tool for code generation, refactoring, debugging, and documentation
• Cursor — Codebase-aware editor, ideal for large projects
• GitHub Copilot — Lightweight everyday completion

Case Studies

Building This Site with Claude Code

Tool: Claude Code Context: Building this Jekyll academic site from scratch — adding pages, tweaking styles, fixing config errors Takeaway: AI grasps Jekyll template structure quickly, cutting down time spent reading docs. Especially effective for niche syntax like Liquid templates.

Debugging Robot Simulation Environments

Tool: Claude Code + Cursor Context: Debugging VLA policies in IsaacSim / MuJoCo, pinpointing sensor data alignment issues Takeaway: Feeding error messages and env configs together to the AI locates issues in minutes that would otherwise take hours.

Paper Writing Assistance

Tool: Claude Context: Polishing Abstract and Related Work sections for the REVER paper Takeaway: AI cannot replace thinking, but it significantly improves language quality. Asking AI to critique first, then revise, works better than asking it to rewrite directly.

Reflections

AI Coding is not about writing less code — it’s about spending your time on decisions that actually matter.

The core skill for engineers using AI tools has shifted to: clearly describing the problem, judging output quality, and keeping the big picture in mind. This maps closely to what robotics research demands — you need to know what the robot should do before you can verify whether it did it right.

Continuously updated. Feel free to reach out by email if a specific use case interests you.