- Building an AI-powered debate platform using Next.js 15 + Claude API, focusing on real-time interaction and structured argument flows.
- Developing a 3D real-time latency visualization system using React Three Fiber, rendering global network data at ~60 FPS.
- Architecting performance-first frontend systems with optimized state management and modular UI design.
- Implementing heavy client-side processing using Web Workers (compression, parsing, async tasks) to reduce server dependency.
The Story
Failure → Shift → Current build.
Act IMediaKit — The Failure
I built MediaKit assuming demand would follow once the product existed. The work was real — privacy-first media tools, solid engineering, things I was proud of technically. But I was building in a vacuum. No validation, no signal, no real user feedback.
After launch: no traction. No adoption. The product existed. The problem it solved, apparently, did not — at least not for anyone I could reach.
What went wrong
- Built before validating problem–market fit
- Assumptions treated as user insight
- No distribution, no audience, no feedback loop
Act IIThe Shift
Quitting MediaKit wasn't the hard part. Understanding why it failed was. The problem wasn't effort — it was sequence. I'd been solving engineering problems when I should have been solving people problems first.
What changed
- Problem-first, not product-first
- Iteration and feedback over completion
- Validation before lines of code
Act IIIDebate AI — Current Build
After MediaKit, I started building Debate AI — an AI-powered debate platform. The early phase was rough: frequent breaking logic, poor prompt outputs, inconsistent responses. Nothing was stable.
I spent roughly two months doing nothing but fixing core logic, iterating prompts, and learning prompt engineering from the ground up — structure, constraints, context control. The kind of depth that only comes from breaking things repeatedly in the same system.
Current state
- System is now functionally stable
- Outputs are consistent and usable
- Built on understanding, not assumptions
Key takeaway
Failure wasn't the issue — wrong approach was. Now I build with validation first, controlled iteration, and a real understanding of what I'm building and why.
