All three of these are multi-agent LLM apps, and they lean on the same trick: agents that communicate through strict JSON-schema contracts rather than free-form prose. That is what makes them reliable enough to chain, route, and revise without a tangle of glue code. Each one is live — switch between them below.
2026 · Design & development
Agentic AI Systems
One showcase, three shipped multi-agent systems — a research pipeline that self-revises, a schema-driven workflow DAG, and a routed wellness chat with shared memory — each demonstrating typed, reliable contracts between LLM agents.
- Next.js
- TypeScript
- OpenAI API
- Structured output
- JSON Schema
- DAG execution
Live demo Multi-Agent Research Pipeline
Researcher → Writer → Critic — each hands the next validated JSON, with an automatic revision loop.
- Three agents hand each other schema-validated JSON instead of prose — zero mapping code between them.
- 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.