Documentation

Short technical write-ups on how Routebook is built — doubles as portfolio depth. Each page is a stub; deeper detail is available on request. The User and Training Guide at the top is the end-user reference for the running app.

User and Training Guide

End-user reference: drawer navigation, map controls, performance overlays, POIs, fitness analytics, and the AI coach.

Architecture overview

How the pieces fit: Next.js App Router, Supabase, Mapbox, AI SDK, MCP, Langfuse. Where TanStack Query ends and Jotai begins.

AI coach: memory, search, tools

Multi-turn coaching with profile and episodic memory, intent routing, hybrid semantic + keyword search, tool-using answers, prompt versioning.

MCP server

Streamable HTTP + SSE, OAuth and Personal Access Tokens, scope-gated tools, resources and prompts, Goose / Claude Desktop wiring.

Fitness model & workout generation

Bannister CTL/ATL/TSB, FTP and LTHR at ride date, power and HR zones, Strava sync or FIT upload, Zwift .zwo generation, training plans.

Case study: productizing this for a coaching business

What a customized deployment looks like — surfaces, scope, integrations, timeline.