Roast My TechStack #2 · April 6, 2026 · Singapore
Three builders brought three very different stacks: a self-improving order-hub workflow, a Spanish-through-music app, and a drag-node automation tool. This recap preserves the practical decisions, trade-offs, and questions raised in the room.
Speaker 01
Harwav Order Hub
Marcus Liang
A self-improving software workflow: users report issues through a chatbot, complaints become GitHub issues, and an automation loop routes them into Claude Code to generate, apply, and ship fixes — with a human validating before redeploy.
Chatbot intake · GitHub issues · OpenClaw · Claude Code · Human validation
01
The Self-Healing Loop
A user complains via the chatbot → the complaint becomes a GitHub issue → an OpenClaw cron pulls the issues → the issue goes into Claude Code → a Claude skill generates and applies a fix → changes push to GitHub → a human tests the software → if solved it redeploys, if not it returns for further fixes.
02
Improving AI Output Around It
To improve AI output quality, check out Hamel's eval blog. And for the UI, shadcn for better-looking components with less effort.
Speaker 02
Kantaro · kantaro.live
Ee One Lim
An app that teaches Spanish through music — pulling real lyrics, front-end design via Stitch, and open-source components to keep the stack light. The roast pushed on where the API calls live and how to keep UI changes verifiable.
LRCLib · Stitch + Google MCP · Heartmula · Open-source streaming · Frontend API calls
01
LYRICS & DESIGN
LRCLib is the open-source package to pull lyrics from music. Use Stitch with Google MCP for frontend design, and Heartmula — an open-source music LLM — as an open-source music streaming / YouTube API alternative.
02
Where The API Calls Live
Push the API call to the frontend to avoid being API-blocked by Google. And add acceptance criteria with screenshots to check for better UI control — the AI should show its work before you approve it.
Speaker 03
Drag Flow
Jesse Lim
An automation tool where you connect workflows by dragging nodes. The roast gave three sharp habits: escalate models deliberately, let AI write your documentation, and make the AI wait for you instead of the other way round.
Drag-and-drop flow · Model escalation · AI documentation
01
Three Habits
- Don't reach for the biggest model for everything — escalate the model accordingly as the task demands it.
- Documentation is your best friend — get the AI to write it for you.
- Let the AI wait for you, instead of waiting for the AI.
Venue sponsor
🏠 Venue Sponsors
An Ting, Raj & SGInnovate
Special shoutouts
🎚 Recording Gears
Jesse Sng
🎙 Mic Runner
Arul