09 — AI fitness app with per-user memory

Athleta

A personal web app for training, nutrition and progress tracking, with an integrated AI agent grounded in application context, memory and a dedicated knowledge base; suggestions are applied only after user confirmation. I handled the full-stack architecture, database modeling, mobile-first UI, AI integration, data validation, deployment and RAG/LLM Wiki systems over data and documentation.

Real interface

01 / 04

Mobile dashboard: active goal, weight, macros for the day and the trend over time

Mobile-first interface with home, training, diet, check-in, AI coach and an admin area.
Active goal with phase, starting weight, current weight and target; weight trend and actual calories against target over time.
Weekly nutrition plan with daily macros and meals followed, skipped or swapped, plus tracked supplementation.
Every user owns an LLM Wiki: profile, goals, rules, baseline, photo analysis and training weeks become a knowledge graph the agent can query.

Stack

ReactTypeScriptViteSupabasePostgreSQLRAGLLM WikiObsidian

In Athleta the AI proposes changes. The user confirms before anything is applied.

1. AI proposes

2. Confirm

—