Agentic Marketplace

Full Project Description

What We Built

The Agentic Marketplace is a generative UI platform where multiple specialized agents operate autonomously, each rendering purpose-built interactive interfaces at runtime. Rather than returning chat messages or markdown, each agent generates and executes rich, interactive UIs—dashboards, scoring systems, code review panels, and approval flows—dynamically tailored to the user’s request.

The Four Agents

📊 Data Analyst

• Ingests raw CSV or JSON data and generates an interactive dashboard in seconds
• Renders Recharts visualizations (bar charts, line graphs, metrics cards) computed on-the-fly
• Calls MCP tools to parse data, compute statistics, and recommend optimal chart types
• Why it’s not a chatbot: The user doesn’t read about their data—they interact with a live, rendered dashboard with filters and drill-down capability

🏆 Hackathon Evaluator

• Scores project submissions across five weighted criteria (innovation, technical execution, UI/UX, business impact, demo quality)
• Generates a radar chart and star-slider UI for real-time scoring feedback
• Presents an approval flow that suspends the agent until the user confirms, requests changes, or escalates
• Why it’s not a chatbot: Evaluation logic is embedded in interactive components; the UI drives the workflow, not the conversation

🔍 Code Reviewer

• Analyzes code snippets for security vulnerabilities and quality issues
• Generates annotated code panels with severity-colored inline annotations (critical, error, warning, info)
• Renders suggestions and an approval flow for accept/request-changes/reject decisions
• Why it’s not a chatbot: Code issues are highlighted visually in context; the user interacts with the review UI, not a text summary

🧭 Orchestrator

• Routes incoming requests to the best-matched specialist agent using LangGraph
• Renders an insight panel showing the routing decision, confidence score, and reasoning
• Navigates the user to the target agent’s workspace with pre-filled context
• Why it’s not a chatbot: Intelligent routing and multi-agent composition happen in the UI layer; the system discovers which agent to invoke dynamically

## Why This Can’t Be Done as a Chatbot

• Data Analyst: A chatbot outputs text summaries. This agent renders an interactive dashboard where users can hover over data points, zoom, and interact with metrics—impossible without UI generation.
• Hackathon Evaluator: A chatbot writes a score in chat. This agent renders a radar chart and slider UI that the user interacts with, then captures their approval decision inline—the approval flow suspends the agent until the user acts.
• Code Reviewer: A chatbot lists issues in markdown. This agent renders color-coded annotations on the actual code, with severity levels embedded in the UI itself.
• Orchestrator: A chatbot says “I think you want the Data Analyst.” This system navigates you there, pre-fills context, and chains agent capability
  together seamlessly.

Key Features

✅ Multi-agent dispatch — Orchestrator intelligently routes to the right specialist
✅ Runtime UI generation — No pre-built dashboards; every UI is generated from data
✅ MCP-powered tools — Agents call structured tools; results drive UI rendering
✅ Approval flows — Inline interactive workflows that suspend and resume the agent
✅ Real-time visualization — Recharts dashboards render with live data
✅ Semantic routing — LangGraph pipeline classifies intent and maps to agents
✅ Generative canvas — /canvas page for building UI layouts in real-time
✅ Component library — /library showcases 35+ component definitions agents can invoke

Why This Matters

Traditional AI systems respond with text. The Agentic Marketplace proves that agents can:

• Design UIs dynamically based on data and context
• Compose multi-tool experiences without hard-coded pages
• Route intelligently between specialists using semantic intent
• Suspend execution for user interaction, then resume with context