VIBE - VELOCITY - AUTOMATION
I build things to think with.
These prototypes are not client brief. They're how I explore ideas, test AI behavior, and figure out what's actually true about a problem.
Production Grade
Project: AI Observability - Claude Code & Design systems
Project: AI Observability
Claude Code & Design systems
Problem : No standard way to observe AI reasoning in production
Solution : Built AI observability dashboard connecting to Langfuse — surfaces traces, evals, and agentic workflows
End Users : AI designers, ML engineers, product teams building on LLMs
Tools : Claude terminal, Langfuse, Vercel. Figma
Agentic Shopping Experience
Project:Beyond Sparky
Project: Beyond Sparky
Problem : Sparky lists. Shoppers need decisions.
Solution : Redesigned Walmart's AI shopping agent with - confidence badges, a named top pick, and a human handoff when the agent isn't sure.
Tools : Claude API · WooCommerce · Cloudflare Worker
Voice Q & A
Project: Retell Supervise
Project: Retell Supervise
I designed Retell Supervise as a system that turns AI agent failures into actionable insights, making agent behavior legible, debuggable, and continuously improvable. Instead of treating QA as review, the product reframes it as a real-time feedback loop where non-technical users can diagnose issues, apply fixes, and validate improvements instantly.
What it does:
- Surfaces and explains why AI voice calls fail (intent, prompt, tool, escalation)
- Maps failures to specific nodes in the agent flow
- Suggests fixes with predicted impact and one-click application
- Enables instant testing through a call simulator (before vs after behavior)
- Identifies patterns across calls to support system-level improvements
Blank Canvas to Design Intelligence
Project: Oneshot Design Copilot
Project: Design Copilot
Designers starting an AI product face the same problem every time. They don't know where to begin. Agent definitions, trust layers, orchestration logic, failure states it's a lot to hold before a single frame is drawn. I built this tool to solve that problem. You describe what you're designing. The system asks the right questions. Thirty seconds later you have a structured AI system blueprint agent definitions, a multi-agent swarm map, trust layer recommendations, and scenario explorations ready to take into Figma.
What it does:
- Zero to AI system blueprint in one session
- Agent definitions + orchestration map
- Trust layer and confidence gating recommendations
- Scenario exploration across edge cases
N8N Automation
Inbox agent
Zero to AI system blueprint in one session. Agent definitions + orchestration map
Zero to AI system blueprint in one session. Agent definitions + orchestration map