DataTalked
Business intelligence platform that transforms Stripe data into actionable insights
Project Overview
DataTalked is a production-ready business intelligence SaaS platform that started as a Shadcn admin dashboard template and evolved into a comprehensive financial analytics solution. Built for the Convex + Resend Hackathon, it combines automated Stripe data synchronization with AI-powered conversational analytics and automated email reporting.
Key Features
RAG-Based AI Chat System
- Multi-Agent Architecture: Manager, Retriever, and Analyst agents working together
- Natural Language Queries: Ask "What's our MRR?" or "Show me revenue trends"
- 7 Structured Business Tools: MRR, revenue, customers, churn, LTV, CAC metrics
- Vector Search: OpenAI embeddings for document retrieval with reranking
- Provenance Tracking: Every answer includes source metadata and confidence scores
Automated Stripe Synchronization
- 3-Tier Sync: Hourly cron job, real-time webhooks, manual refresh
- Comprehensive Data: Payments, customers, subscriptions, disputes, refunds
- Background Processing: All data cached in Convex for fast queries
- Webhook Security: Signature verification and event ID logging
Real-time Analytics Dashboard
- Financial Metrics: Revenue, MRR, churn rate, LTV, CAC with growth indicators
- Interactive Charts: Revenue trends, customer acquisition, product performance
- Demo Data Mode: 580+ days of realistic financial data for testing
Automated Email Reports
- Flexible Scheduling: Daily, weekly, monthly, quarterly reports
- Beautiful Templates: React Email components with comprehensive metrics
- Smart Scheduling: 30-minute precision with timezone support
- Delivery Tracking: Email history with open rate monitoring via Resend
Tech Stack
Frontend
- React 19 with TypeScript (strict mode)
- TanStack Router (file-based routing)
- ShadcnUI (TailwindCSS v4 + RadixUI)
- Recharts for data visualization
- Vite for fast development builds
Backend & Database
- Convex (real-time serverless database)
- Convex Actions for external API calls
- Convex Crons for scheduled tasks
- HTTP endpoints for webhooks
- Stripe API integration
AI & Integrations
- OpenAI GPT-4 (chat and embeddings)
- Anthropic Claude (alternative model)
- Convex AI Agent framework
- Convex RAG package for vector search
- Resend for email delivery
Development
- Bun for package management
- ESLint & Prettier for code quality
- GitHub Actions for CI/CD
- Netlify for frontend hosting
- Convex Cloud for backend
Technical Highlights
AI Chat Architecture
The chat system uses a multi-agent approach where the Manager Agent analyzes query complexity, the Retriever Agent searches relevant documents using RAG, and the Analyst Agent executes structured business metric tools. All responses include provenance tracking with source citations and confidence scores.
Stripe Sync Strategy
Instead of making live API calls, DataTalked implements a three-tier synchronization system: hourly background cron jobs fetch all data, real-time webhooks handle instant updates, and users can manually refresh when needed.
Demo Data System
The platform includes a comprehensive demo mode with 580+ days of realistic financial data (Jan 2024 - Aug 2025) generated using Faker.js. This enables users to explore features without connecting their Stripe account.