The Brief
An entrepreneur wanted to test the waters in the AI-powered entertainment niche - specifically, tarot card readings. The vision wasn't a chatbot with a tarot skin. It was a full interactive experience: multiple reader characters with distinct personalities and voices, cards drawn one at a time with dramatic pacing, spreads that animate and unfold, and an AI that remembers you across sessions.
What We Built
A monorepo with a client app, an AI agent server, an admin dashboard, and shared packages for the database, types, and configuration.
AI-Controlled Interface
The core of the product: the AI agent doesn't just generate text - it controls the visual experience. Through a set of custom tools built on the Mastra agent framework, the agent selects spreads, draws cards, reveals positions, manages the reading flow, and responds to user input - all reflected in the UI in real time through streaming updates.
The reading unfolds in phases. The agent guides the user through setup, draws cards one at a time with pauses for reflection, interprets the full spread, and closes the session. An auto-continue timer keeps the pace moving between draws, creating the dramatic tension of a real reading.
Multiple Reader Characters
Each reader is a distinct AI persona with its own personality, archetype, reading style, and voice. The voices are generated with ElevenLabs TTS, so each character actually speaks. New characters can be added from the admin dashboard without touching code.
Persistent Memory
The agent remembers users across sessions. A vector-based memory system stores context from past conversations - preferences, personal details, themes from previous readings - and retrieves relevant memories to personalize each new session.
Full Tarot Content System
A complete 78-card deck with upright and reversed meanings, multiple spread types with named positions, and categorized question prompts that help users formulate focused questions. All content is database-driven and manageable through the admin interface.
The Result
This project demonstrates what happens when you build a purpose-specific AI experience instead of wrapping a chatbot in a theme. The architecture - where an AI agent orchestrates the entire UI through tool calls - is a pattern that applies far beyond tarot: any product where AI needs to do more than return text.