DataSense
Team led by a SPURT Product Developer Intern skilled in Flutter, Django, and Angular, building open-source identity protocols (DIDs) for AI agents and MCP.
YouTube Video
Project Description
DataSense — Kill the Dashboard
“Agentic Interfaces — build UIs that agents render at runtime.”
DataSense is a generative, agent-driven data visualization platform built for the Generative UI Global Hackathon. It is designed to replace static dashboard templates with a dynamic, LLM-orchestrated experience where the interface is a direct consequence of the data analysis.
🎯 The Vision: “Kill the Dashboard”
Traditional dashboards are rigid, pre-built, and often overwhelming. DataSense follows the “Kill the Dashboard” hackathon track:
No Pre-built Pages: The entire UI is generated at runtime based on the uploaded data.
Agent as Architect: The agent decides the layout (Bento Box), the chart types, and the analytical focus.
Beyond the Chat Bubble: It provides a spatial, interactive workspace that a text-only chatbot simply cannot replicate.
🛠️ The Tech Stack
Built on the core protocols of the Generative UI frontier:
🧠 Brain: LangChain + Google Gemini 2.5 Flash.
📐 UI Protocol: A2UI (Agent-to-User Interface) — Declarative, client-agnostic component blueprints.
🔄 Sync Layer: AG-UI / CopilotKit — Real-time state synchronization between Agent and UI.
🔌 Connectivity: MCP (Model Context Protocol) — Backend exposed as a functional MCP Server for cross-agent tool usage.
🎨 Design: Terminal-inspired “Hyperstudio” aesthetic with an Elastic Bento Box layout.
🚀 How It Works
Ingest: Data is analyzed via a MCP-compliant profiling engine.
Reason: A Gemini agent analyzes the data patterns (Temporal, Categorical, Correlational).
Orchestrate: The agent generates an A2UI Adjacency List specifying the exact tiles needed.
Render: The React frontend synchronizes via AG-UI, rendering a responsive, drag-and-drop dashboard.