Shipped: IceBreak 🧊

Built a tool I actually needed — AI-generated conversation starters before remote work calls, tuned by your colleague's country, age group, and meeting type.

The problem: working across Germany, Japan, and the US from India means small talk that feels natural to me often falls flat on the other side. Generic openers are worse than silence. And I don't want to prompt an LLM for 2 minutes when I have 30 seconds before a call starts.

How it works:
→ Pick 3 things about your colleague (country / age / meeting type)
→ Optionally add a personal detail — the model prioritises it
→ Get 3 culturally-aware openers, revealed one by one as flip cards
→ See what cultural reference Claude drew from ("Based on →")
→ Copy or share any starter directly from the card

Features:

• 20 countries with search

• Cultural hook — Claude picks one timely, non-political reference for that country right now. No generic openers, no out-of-season festivals.

• Copy-to-clipboard + shareable links on every card

• No sign up, no prompting, 30 seconds

Tech:

• React + Vite, plain CSS — no component library

• Vercel serverless function calling Claude via raw fetch (not the SDK — it silently fails in Vercel's runtime, cost me 3 hours)

• Google Sheets email capture via Apps Script webhook

• In-memory rate limiting (10 req/IP/hour)

🔗 https://icebreaker-tau-rouge.vercel.app
👨💻 https://github.com/aruna09/icebreaker

Would love feedback — especially on whether the cultural hook actually makes the outputs feel less generic.