AI-First Community Racquet Game Booking

Product challenge

Community racquet-sport players often coordinate across group chats, venue directories, and booking systems. The product challenge was not simply to add a chatbot, but to understand this fragmented journey and make hosting, discovering, and joining a game feel coherent without hiding important decisions from the player.

Product and engineering approach

Start with the player journey

• Built a conversational assistant that turns natural-language requests into structured match and venue actions.
• Created guided workflows for searching games, hosting a match, joining an open seat, updating plans, and cancelling when needed.
• Kept user intent and confirmation central to consequential actions, while using player profiles and preferences to improve relevance.

Make AI a product interface, not a separate demo

• Designed an authenticated MCP service exposing match, venue, and player operations as reusable tools.
• Shared Zod contracts across the web application, agent, MCP service, and domain layer to keep tool calls type-safe.
• Implemented OAuth-aware request handling so each action runs in the authenticated player's context.

Build a domain that can grow

• Built venue resolution with local records, semantic ranking, and Google Places support.
• Added domain workflows for open seats, reserved players, match moderation, and Mexicano-style rotations.
• Structured the product as a pnpm monorepo with separate Next.js, LangGraph, MCP, and shared-package workspaces.

Outcome

• Consolidated venue discovery, match coordination, and player management into one AI-assisted product journey.
• Made the same trusted domain capabilities available to the web experience and external AI clients through a shared MCP surface.
• Established a modular product foundation for learning from real community workflows and adding new sports, formats, and booking integrations.