Automation
AutoFit AI Coaching System
The problem wasn't sending messages. It was the time spent preparing them.
Summary
- BuiltDesigned an immutable weekly snapshot system — each client's progress is preserved as a discrete record rather than overwritten, creating a reliable coaching history.
- ProblemA fitness coaching business had a preparation problem, not a messaging problem.
- ImpactCoach no longer starts every weekly message from zero — AI handles progress analysis and drafting, coach focuses on judgment, tone, and the final call.
Problem
A fitness coaching business had a preparation problem, not a messaging problem.
A fitness coaching business had a preparation problem, not a messaging problem. Every week, the coach manually reviewed each client's progress, compared it with prior weeks, and wrote personalized updates from scratch.
As the client roster grew, this became the primary bottleneck. The coach's real value is judgment and relationships —
not repetitive analysis that a system could handle. The goal was to remove the prep work without removing the human quality.
What I Built
5 key contributions that defined the build
- Designed an immutable weekly snapshot system — each client's progress is preserved as a discrete record rather than overwritten, creating a reliable coaching history.
- Built the AI analysis layer that compares previous and current progress data, then generates a draft coaching message with observations, recommendations, and motivational framing.
- Implemented the coach approval gate — every generated message requires review before delivery, making human judgment the final authority.
- Created a regeneration loop: if the coach requests edits, the system refines the draft using their feedback rather than starting from scratch.
- Engineered prompts for tone consistency — messages needed to sound like the coach, not like an AI summary.
Key Decisions
Tradeoffs and approaches that shaped the final outcome
Challenge
Trust-sensitive content fails at both extremes — messages that feel robotic destroy the coach-client relationship even if technically accurate.
Approach
Designed the approval gate as the core of the system architecture, not an optional add-on — automation becomes leverage only when the human remains in control.
Challenge
Overwriting client progress state between weeks would break the AI's ability to reason about behavioral trends over time.
Approach
Used immutable weekly snapshots so the AI always reasons from preserved observation history, not a destructively updated state.
Impact
Coach no longer starts every weekly message from zero — AI handles progress analysis and drafting, coach focuses on judgment, tone, and the final call.
- Coach no longer starts every weekly message from zero — AI handles progress analysis and drafting, coach focuses on judgment, tone, and the final call.
- Improved message consistency across the client roster and reduced weekly preparation time without reducing personalization quality.