WhyRep: Constraining LLM coaching to validated exercise science methodology

Short summary

A founder describes building WhyRep, a workout tracker with an AI coach that uses Claude only as a conversational layer constrained by pre-written, validated exercise science methodology. Instead of fine-tuning or letting the LLM generate advice freely, deterministic engines implement progression, autoregulation, and plateau diagnosis logic from authored documents with test vectors. The LLM explains decisions in plain language but cannot invent new training science, ensuring every recommendation is scientifically justifiable.

• •WhyRep uses a doc-first approach: methodology is written and validated before any coaching code
• •Deterministic engines implement the methodology; Claude handles only the conversational explanation layer
• •Every AI recommendation must trace back to approved, testable rules — not LLM-generated science

Generated with AI, which can make mistakes.