Live

AI Cycling Coach

AI-powered personalized cycling training

ai-cycling-coach
ai-cycling-coach

The Problem

📋

Generic Training Plans

Off-the-shelf training plans ignore individual fitness levels, recovery status, and personal goals. Athletes follow cookie-cutter programs that don't adapt to their actual performance data or daily readiness.

No Power Zone Analysis

Most cyclists train without understanding their power zones, leading to inefficient workouts that either undertrain or overtrain specific energy systems. Without zone-based insights, training stimulus is suboptimal.

📝

Manual Workout Planning

Creating structured workouts with proper intervals, rest periods, and progressive overload requires deep coaching knowledge. Athletes spend hours designing workouts that a coach could prescribe in minutes.

Before

Generic training plans from blogs and YouTube videos

After

AI-generated workouts personalized to your FTP, TSB, and daily readiness

Before

Manual conversion of workout ideas to Zwift-compatible format

After

Instant ZWO file export — ride the AI-designed workout immediately on Zwift

Before

Guessing training intensity without objective data analysis

After

TSB-based intensity filtering ensures optimal training stimulus every session

Approach

Omakase Pattern: Curated AI Selection

Instead of generating workouts from scratch (which risks producing physiologically unsound sessions), the system uses an 'Omakase' pattern — pre-validated workout modules designed by coaching science are stored in a library, and the AI selects and assembles the optimal combination based on your current TSB (Training Stress Balance), FTP, weight, and wellness score. This ensures every workout is both scientifically valid and personally optimized. The output is a Zwift-compatible ZWO file you can ride immediately.

Define
Execute
Collect
Report

BDD Pipeline Flow

Key Features

ai-workout-generator
ai-workout-generator

AI Workout Generator

Select your training goal and available time, and the AI assembles a structured workout from pre-validated modules. Uses LLM intelligence to match workout intensity to your current fatigue level (TSB) and training phase, outputting a ride-ready ZWO file.

power-zone-analysis
power-zone-analysis

Power Zone Analysis

Automatically calculates and visualizes your 7 power zones based on FTP. Shows time-in-zone distribution across workouts, identifies training gaps, and recommends zone-specific sessions to build a balanced fitness profile.

training-calendar
training-calendar

Training Calendar

Weekly periodization view showing planned vs. completed workouts, TSS (Training Stress Score) targets, and rest day recommendations. Syncs with Intervals.icu for automatic training data import and progress tracking.

performance-tracking
performance-tracking

Performance Tracking

Long-term performance trends with FTP progression, CTL/ATL/TSB charts, and power curve analysis. Tracks improvements across training blocks and provides data-driven insights for peak performance timing.

Architecture

SPAREST APILLM PromptOAuth / DataSQL
Browser
Next.js Frontend
Python / FastAPI
Claude API
Strava API
Supabase
Client
Server
Database
Service
External

Results

7
Power Zones

FTP-based zone calculation for personalized training

TSB
Readiness Score

Training Stress Balance for daily intensity guidance

ZWO
Instant Export

Zwift-compatible workout files generated on demand

Weekly
Periodization

AI-planned weekly training structure with recovery days