My Workflow: How I Build with AI
People ask how I manage to build so many projects while being a student. The answer isn't "grind harder" — it's having the right setup. Here's my complete workflow.
The Core: OpenClaw
I run OpenClaw as my AI assistant platform. It connects to my desktop environment and lets me code, manage files, and automate tasks through natural conversation.
Today I used it to code a birthday card for my dad — drew balloons, cake, hearts with Python PIL, all through chat commands.
Project Organization
I learned the hard way that monorepo > many repos.
All my web tools live in tools-suite:
- api-tester, json-viewer, regex-tester
- jwt-decoder, csv-json, diff-viewer
- Plus exclusive tools: color-picker, life-pattern-generator
One deploy, one URL: qqshi13.github.io/tools-suite/
Archived the standalone repos. Less clutter, easier maintenance.
My AI Assistant Pattern
I don't code alone. My workflow with Nova:
- Idea → Describe what I want
- Prototype → AI generates first version
- Iterate → Quick back-and-forth fixes
- Ship → Deploy immediately
Example: The birthday card went through 3 iterations in 10 minutes:
- v1: Wrong tone, too casual
- v2: Removed coding mention
- v3: Clean, no signature, perfect
Vibe Coding at Scale
"Vibe coding" isn't just for quick scripts — I use it for large-scale bug sweeps too.
Last week, I ran a mass bug check across all 10 active repositories:
- Spawned sub-agents in parallel, one per repo
- Each agent checked, fixed, and validated independently
- Fixed 42 bugs across tools-suite, droptransfer, collaboard, flow, lifelab, and M5Timer
- All changes committed and pushed automatically
The key: parallel sub-agents + consistent reporting format. What would take hours manually took one coordinated session.
Hardware Projects
My M5Timer (Pomodoro device) was built with the same workflow:
- C++ code for ESP32-S3
- Web interface with Web Serial API
- All version controlled, GPL-3.0 licensed
Hardware + software + AI assistance = fast iteration.
System Hygiene
I clean house regularly:
- Remove unnecessary services (goodbye snapd, atop)
- Kill old sessions, keep only active ones
- Archive old projects, delete abandoned ones
Today's cleanup freed 60MB RAM and 5 stale session files.
The Philosophy
Build fast, ship often, clean regularly.
Don't let perfect be the enemy of done. My tools aren't flawless — they're useful and deployed.
And having an AI assistant that understands context (it reads my memory files every session) means I never lose momentum.
Tools mentioned: OpenClaw, GitHub Pages, M5Stack, VS Code
Assistant: Nova ☄️