ahad.

Home/Projects/AI Gym Memory System

AI Gym Memory System

Conversational workout tracker with sub-second intent extraction via Gemini Flash. Log exercises in natural language and query history semantically — 'What did I train last Tuesday?' — with ChromaDB vector retrieval.

PythonFastAPIGemini FlashChromaDBPostgreSQL
Source Code

Overview

Traditional workout tracking apps require tedious manual input. This system lets users log workouts conversationally:

  • "I did 3 sets of bench press with 185 lbs today"
  • "What did I train last Tuesday?"
  • "Show me my shoulder workouts from this month"

Architecture

User Input → FastAPI Backend → Gemini Flash (Intent Extraction) → Activity Parser (exercise, reps, sets, weight, date) → Embedding Service → Storage Layer (PostgreSQL + ChromaDB)

Retrieval uses a hybrid approach combining semantic similarity, keyword matching, and temporal filtering.

Technical Highlights

  • Intent Extraction with Gemini Flash: Extracts structured data from unstructured natural language, handling ambiguity and context
  • Dual Storage: PostgreSQL for structured/transactional data, ChromaDB for fast vector similarity search
  • Hybrid Query: Combines semantic search, keyword matching, and temporal filtering with configurable similarity thresholds
  • Metadata Indexing: Efficient retrieval by date, exercise type, and muscle group