The main dashboard of DebateMe
The coin toss
The UI of me vs ai showing how ai is responding to my arguements

What Inspired Me

I've always believed that the best way to sharpen your thinking is to have someone push back hard against it. But most AI tools are designed to agree with you — to assist, to validate, to make things easier.

I wanted to build the opposite. An AI that argues back. An AI that never concedes. An AI that gets more ruthless the better you get.

DebateMe was born from a simple question: what if your AI opponent actually wanted to win?

How I Built It

I built DebateMe entirely using MeDo — Baidu's natural language app builder — without writing a single line of traditional code. Every feature, every behavior, every animation was described in plain English and generated by MeDo.

The build happened in stages:

Stage 1 — The Debate Engine The core challenge was making the AI behave like a real debate opponent, not a chatbot. I structured strict behavioral rules through MeDo's conversation system:

The AI must directly reference the user's exact words in every response
It must never repeat a counter-argument already made in the same session
It escalates in intensity across 6 turns — weakest argument first, strongest last
It always ends with a sharp challenge or question

Stage 2 — Difficulty Scaling Three modes required three completely different AI personality profiles:

Casual: broad arguments, conversational tone
Competitive: real statistics, assertive style
Brutal: names logical fallacies explicitly, cites authoritative sources, zero mercy

Stage 3 — The Features

Dramatic coin toss system to assign FOR/AGAINST positions with a 3D flip animation
Real-time fallacy detection banner in Brutal mode
Live scoring system (Clarity, Evidence, Logic) with animated progress bars
Voice input via Web Speech API
Full verdict page with AI judge analysis

The Biggest Challenge

The hardest problem I faced was AI repetition.

In early testing, the AI was giving the same 4 responses regardless of what the user argued. It had fallen into a template loop — inserting the user's words into the same sentence structure every time.

The fix wasn't code. It was prompt engineering.

I restructured the AI's instructions to explicitly require it to:

Reference the user's specific argument before countering
Track what it had already said and never repeat it
Make substantive topic arguments, not word attacks
Escalate progressively rather than staying flat

One well-structured prompt solved a problem that would have taken hours to fix in traditional code.

That's the power of building with MeDo.

What I Learned

Prompt engineering is software engineering — precision in language produces precision in behavior
The most impressive features aren't always the most complex to build — fallacy detection took one clear instruction to MeDo
Scope discipline wins hackathons — a polished focused app beats an ambitious unfinished one
Building without code doesn't mean building without craft — the craft is in how clearly you think and communicate