divyesh-1729/fitpilot

Purpose and Entry Points

This repository implements a Streamlit-based workout coaching application: it analyzes exercise video/pose data, counts or evaluates repetitions, tracks workout metrics, persists records, and can provide text/voice coaching. The primary application entry point is main.py; Streamlit behavior and theme settings are rooted at .streamlit/config.toml. A separate static landing page is provided at LandingPage/index.html.

Application Composition

The Streamlit UI orchestrates authentication, workout selection, exercise processing, metrics display, and coaching interactions. Shared UI presentation is handled separately from core workout logic:

• main.py

• services/auth/login_wall.py

• services/ui/style_loader.py

• static/style.css

Exercise Domain and Detection Layer

Exercise-specific detectors encapsulate pose-based form/state logic for supported movements. A base abstraction likely defines the contract shared by detector implementations, separating generic exercise lifecycle/state handling from movement-specific rules:

• core/base_exercise.py

• detectors/biceps_curl.py

• detectors/lunges.py

• detectors/pushup.py

• detectors/shoulder_press.py

• detectors/squat.py

Vision Processing Boundary

Video ingestion and pose inference are isolated in the vision service. This boundary should transform frames/video into pose landmarks and detector-ready observations, keeping MediaPipe/model concerns out of Streamlit pages and exercise classes. The packaged pose model is a local runtime asset rather than a remotely hosted inference dependency:

• services/vision/exercise_video_processor.py

• ml_models/pose_landmarker_full.task

• detectors/__init__.py

Session, Workout, and Metrics Flow

At startup, session defaults initialize Streamlit state. Workout configuration selects exercise behavior/settings; processed exercise events feed metrics tracking; resulting summaries are available to UI, persistence, and coaching stages:

• services/state/session_defaults.py

• services/config/workout_config.py

• services/tracking/metrics.py

• core/base_exercise.py

Persistence and Coaching Services

Workout/exercise data is persisted through a repository boundary backed by the included SQLite database. Coaching is modularized into LLM text generation, text-to-speech, and a voice pipeline, allowing coaching outputs to remain downstream of workout analysis rather than embedded in detector code:

• services/persistence/exercise_repository.py

• data.db

• services/coaching/llm.py

• services/coaching/tts.py

• services/coaching/voice_pipeline.py

Runtime and External Dependencies

The runtime is Python with Streamlit, configured through requirements.txt and packages.txt. Material local infrastructure includes SQLite (data.db), the pose-landmarker model (ml_models/pose_landmarker_full.task), and static/font assets. The LLM and TTS modules indicate external AI/voice service integration, though provider credentials and deployment configuration are not present in the repository tree.