Skip to content
Andrew VanDykeAndrew VanDyke

2026 · Design & development

Agentic AI Systems

Three small multi-agent systems: a research pipeline that revises itself, a typed workflow DAG, and a wellness chat that routes requests while sharing context.

  • Next.js
  • TypeScript
  • OpenAI API
  • Structured output
  • JSON Schema
  • DAG execution
Three live demos

All three are multi-agent LLM apps built around the same idea: agents communicate through strict JSON-schema contracts instead of free-form prose. That keeps the hand-offs easier to chain, route, and revise. You can switch between the running demos below.

Research pipeline UI showing structured findings and a drafted article on walk-forward optimizationLive demo

Multi-Agent Research Pipeline

A researcher, writer, and critic pass validated JSON through an automatic revision loop.

  • Three agents pass schema-validated JSON instead of prose, so the hand-offs need no mapping code.
  • The Critic's structured feedback triggers an automatic revision pass before the article is finalized.
  • Every step is typed and inspectable: watch findings become a draft, get critiqued, and get rewritten.
Open live demo Source