Tableau AI Copilot

Landing Page
Copilot Chat Page
Querying
Visualization Generation for the previous query
Canvas Tab to pin your visualizations for later reference

Inspiration

The inspiration for Tableau AI Copilot came from observing a common pain point: finding the right visualization in Tableau Cloud is harder than it should be. Users have hundreds of dashboards but don't know what exists, can't search effectively by name, and waste time browsing manually. Non-technical users struggle even more.

We envisioned a future where you could simply ask "Show me Q4 sales performance" and instantly get the right dashboard—no browsing, no guessing, just natural conversation with your data. That's why we built Tableau AI Copilot: to democratize access to analytics through the power of AI.

What it does

Tableau AI Copilot transforms how you interact with Tableau Cloud by adding an intelligent, conversational layer powered by Google Gemini AI and LangGraph agents.

Key Features: • Natural Language Discovery: Ask "Show me sales dashboard" and AI semantically searches all Tableau content • Embedded Visualizations: Dashboards appear instantly in chat with full interactivity • Secure Authentication: JWT-based authentication using Tableau Connected Apps • Multi-Agent AI System: Router, Discovery, Retriever, Analyzer, and Summarizer agents working together • Real-time Progress: Watch agents work with animated status indicators

Tech Integration: • Tableau REST API v3.22 for metadata • Tableau Embedding API v3 for interactive dashboards • OpenAI embeddings for semantic search (50-60% relevance) • Google Gemini Pro for reasoning • ChromaDB for vector storage

User Journey:

Connect to Tableau Cloud (one-click authentication)
Workbooks automatically indexed with AI embeddings
Ask questions in natural language
AI finds relevant dashboards and displays them inline
Interact with live Tableau visualizations directly in chat

How we built it

Architecture: React + TypeScript frontend ↔ Node.js backend ↔ LangGraph agents ↔ Tableau Cloud + Google Gemini + OpenAI

Frontend: • React 18 with TypeScript for type safety • Tailwind CSS and shadcn/ui for modern UI • Tableau Embedding API v3 for visualizations • Socket.IO for real-time updates

Backend: • Node.js + Express with TypeScript • LangChain.js + LangGraph for multi-agent orchestration • Google Gemini Pro for reasoning • OpenAI embeddings for semantic search • Tableau REST API v3.22 integration • JWT authentication with Connected Apps • ChromaDB for vector storage

Key Implementation:

Tableau Integration: JWT token generation, REST API client with 12+ endpoints, semantic indexing
AI Agents: State-based LangGraph workflow with 5 specialized agents and conditional routing
Semantic Search: OpenAI embeddings (1536-dim vectors), cosine similarity ranking
Security: Server-side credentials, session-based auth, CORS protection

Development took [X weeks] with comprehensive testing at each phase.

Challenges we ran into

Tableau Authentication: Connected Apps JWT authentication required understanding Direct Trust, secret signing, and token lifecycle. Solution: Created proper JWT generation with automatic refresh and comprehensive test scripts.
Semantic Search Accuracy: Initial keyword matching returned irrelevant results. Solution: Switched to OpenAI embeddings with rich metadata descriptions, achieving 50-60% relevance.
Embedding API Integration: Tableau's Web Components required special React handling and JWT refresh. Solution: Built React wrapper component managing lifecycle and authentication.
Multi-Agent Orchestration: Coordinating multiple AI agents while maintaining state was complex. Solution: Used LangGraph's StateGraph with explicit transitions and WebSocket progress updates.
Real-Time Updates: Users needed to see progress during 5-15 second AI processing. Solution: Implemented Socket.IO with animated agent status indicators.
TypeScript Types: Integrating multiple libraries with proper types required significant effort. Solution: Created comprehensive type definitions for all APIs and state schemas.

Accomplishments that we're proud of

Technical Achievements: • Full Tableau Platform integration (REST API v3.22 + Embedding API v3) • Advanced multi-agent AI system with LangGraph • Semantic search with 50-60% relevance accuracy • Production-ready TypeScript codebase with 15,000+ lines • 30+ API endpoints, 12+ Tableau integrations

Innovation Highlights: • First AI copilot specifically for Tableau Cloud • Semantic discovery using embeddings (not keyword search) • Embedded chat experience with inline visualizations • Multi-agent intelligence with specialized tasks

Impact Potential: • Reduces dashboard discovery from minutes to seconds • Makes BI tools accessible to non-technical users • Helps users discover visualizations they didn't know existed • Increases Tableau ROI through better discoverability

What we learned

Technical Learnings: • Tableau Developer Platform: Deep dive into Connected Apps (JWT + Direct Trust), REST API v3.22, Embedding API v3 architecture • AI Orchestration: LangGraph's StateGraph for complex agent workflows, conditional routing, state management • Semantic Search: OpenAI embeddings capture meaning better than keywords, quality metadata is crucial • Full-Stack TypeScript: End-to-end type safety catches errors early and makes refactoring safer • Real-Time UX: WebSocket progress updates transform user perception of AI speed

Product Learnings: • Real-time feedback is essential for AI applications • Server-side credentials are simpler than OAuth flows for users • Showing agent progress builds trust in AI results • Natural language lowers barriers to entry significantly

Key Lessons:

Start with authentication (everything depends on it)
Test each component independently with scripts
TypeScript from day one saves maintenance time
Design state schema before building agents
AI quality depends heavily on prompt engineering

What's next for Tableau AI Copilot

Immediate Roadmap: • Tableau Pulse integration for metric insights • Custom view generation based on natural language • Multi-workbook analysis and comparison • Slack and Teams bot integration

Long-Term Vision: • Enterprise features: Multi-tenancy, SSO, audit logging • Platform expansion: Mobile apps, voice interface • AI enhancement: Fine-tuned models, reasoning chains • Data sources: Direct DB connections, cloud warehouses, SaaS integrations