Artifex

Problem: Building multi-agent AI systems requires deep technical knowledge of agent frameworks, architecture design, and Python code generation — a complex, time-consuming process inaccessible to most developers without specialist expertise.

Solution: Built Artifex, a next-generation AI agent builder that converts natural language requirements into complete, production-ready multi-agent Python systems using Google ADK, featuring a FastAPI/WebSocket backend, a visual Next.js frontend with real-time graph visualization, and a meta-agent orchestrator that runs sub-agents for requirements analysis, architecture planning, tool generation, and code output.

Key Features:

• 🧠 Meta-Agent Orchestration: Google ADK-powered orchestrator chains specialized sub-agents (Requirements Analyzer → Architecture Planner → Agent Builder → Tool Builder → Code Generator → Validator) to convert free-text into executable Python projects
• 📊 Visual Agent Graph: Real-time ReactFlow architecture visualization streamed over WebSocket — shows agent nodes, tool connections, and sub-agent hierarchies as they are built
• 🔧 Custom Tool Generation: Generates Python tool functions with error handling for Google Search, URL context, API integrations, data processing, and file I/O
• 💻 Integrated Code Editor: Syntax-highlighted TipTap editor for previewing and editing the generated agent.py, tools.py, requirements.txt, and README.md inline
• 🔄 4 Agent Type Support: Generates LLM Agent, Sequential Agent, Parallel Agent, and Loop Agent configurations with correct ADK wiring
• 🚀 Session Management: Persistent config state across the full creation conversation — requirements, architecture, tool specs, and generated files survive multi-turn sessions

Technical Implementation:

• Built meta-agent system with Google ADK, coordinating 6 specialized sub-agents via a root orchestrator defined in Pydantic config schemas
• Designed FastAPI REST + WebSocket API with live progress streaming to the frontend during code generation
• Built Next.js frontend with ReactFlow for interactive agent graph visualization, TipTap for code editing, and Tailwind CSS + shadcn/ui for the interface
• Implemented config-to-code pipeline converting validated Pydantic agent configs into complete Python ADK project files
• Supported 8+ builtin tools (google_search, url_context, load_memory, transfer_to_agent, exit_loop, etc.) and arbitrary custom Python tool generation
• Auto-detected monorepo structures and generated project-ready output with requirements.txt and README.md for immediate use

Impact: Delivered a full-stack AI agent builder that collapses multi-agent system development from hours of manual coding into a single natural language conversation — producing complete, runnable Python ADK projects with visual architecture previews and integrated editing.