Altio
An AI coach that cites your own data
Problem
Health data is scattered, and no app tells you what to do with the numbers. I wanted a coach that sees everything — sleep, sessions, food, journal — and backs every recommendation with the user’s own data, not generic internet wisdom.
Solution
Two-tier coaching
A deterministic progression engine computes next session’s loads instantly and for free (RPE × reps → +2.5 kg / hold / -10% deload). The weekly LLM review takes that output as input and may override it based on recovery data. The boring math stays auditable — the AI adds judgment.
Memory that consolidates overnight
Every night Claude summarizes the week’s logs into typed memories (training / recovery / mindset) in pgvector, with source references. The coach retrieves them by cosine search and always shows which numbers a recommendation rests on.
Offline first
Every log entry goes through a write queue: failed saves retry with exponential backoff, and terminal failures land on a user-visible dead-letter list. A basement gym breaks nothing.
A statistical insight miner
Pearson correlations over a 60-day window: sleep vs. next-day RPE, HRV vs. tonnage. An insight only shows when n ≥ 10 and |r| ≥ 0.25 — no numerology.
Gallery
