Deven Varu

Agent infrastructure / Jan 2026 - Mar 2026

A2A Multi-Agent Builder

Platform for creating, versioning, and orchestrating multi-agent AI systems dynamically instead of hardcoding workflows.

Independent prototype / Full-stack AI engineer / Prototype

A2A / TypeScript / MongoDB Atlas / Prompt Versioning / Agent Orchestration

A2A Multi-Agent Builder Project screenshot coming soon

Highlights

  • Designed an agent factory that assembles specialized teams at runtime.
  • Supported parallel and sequential execution with structured response aggregation.
  • Used MongoDB Atlas for prompts, agent configs, routing logic, shared context, execution logs, versioning, and rollback.

Problem

Hardcoded agent workflows make it difficult to create, adjust, version, and roll back specialized multi-agent teams.

What I built

  • An agent factory that assembles specialized teams at runtime.
  • Parallel and sequential execution modes with structured response aggregation.
  • MongoDB Atlas storage for prompts, agent configs, routing logic, shared context, execution logs, versioning, and rollback.

Technical focus

The project centered on dynamic agent configuration, prompt versioning, shared context, and execution observability.

View external project Back to projects

Related projects

  • Codex Session Visualizer - VS Code extension that visualizes how Codex explores, reads, edits, creates, and deletes files during AI coding sessions.
  • 10-Agent AI Board Game Simulation - Hierarchical multi-agent system where autonomous AI characters play a custom board game through movement, dialogue, actions, and combat.
  • InterviewWithAI - AI-powered interview practice platform for realistic voice interviews, coding problems, transcripts, and structured performance feedback.