PoolPilot - AI-Driven Athletic Orchestration

Technical Overview

PoolPilot is an intelligent generation platform for aquatic training protocols. It utilizes Google Gemini 2.5 Flash to transform high-level user parameters into structured, executable workout data based on modern sports science. The engine implements sophisticated prompt engineering using 4-dimensional parameter mapping (Focus, Level, Style, Zone) and enforces strict JSON schema compliance for AI outputs. Designed as a stateless serverless application, it handles complex athletic periodization for swimmers ranging from Beginner to Olympic levels.

Problem Statement

Swimmers lack access to personalized, scientifically-backed training plans. Generic routines fail to adapt to individual energy levels, technical focus, and competition schedules.

Architecture

Serverless-ready Next.js (App Router) application. Workflow: Parameters -> Auth Check -> Dynamic Prompt Construction -> LLM Inference -> Schema Validation -> UI Hydration. Includes a GenerationLimitService for tier-based access control.

Key Features

• LLM Orchestration with Google Gemini 2.5 Flash
• 4-Dimensional Parameter Mapping (Focus x Level x Style x Zone)
• Structured JSON Schema Validation for AI Outputs
• Serverless-Ready Stateless Architecture
• On-the-fly Dynamic PNG Asset Generation
• Tier-based Entitlement Logic (Free vs. Pro)

Challenges

• Ensuring strict schema compliance for diverse LLM-generated routines
• Constructing effective context-aware prompts for high-performance athletes
• Managing stateful generation limits in a stateless environment
• Implementing high-fidelity server-side canvas rendering for training exports

Outcomes

Generated over 10,000 personalized training plans with high user satisfaction. Successfully implemented a robust limit system and a tournament mode with specific periodization.