How twarita11/ayurvedic-etongue-ml works
Project type and purpose
• This is a specialized full-stack ML pipeline for analyzing NIR sensor data from Ayurvedic medicines.
• Purpose: preprocess raw sensor CSVs, engineer features, train/serve an ensemble of ML models (RF, SVM, NN), and expose REST endpoints for dilution, medicine classification, and effectiveness estimation. There is also a visualization layer (notebooks and real-time plots).
File structure
• Top level:
• data/ – stores raw & processed CSV data
• models/ – serialized model artifacts (pickles, .joblib)
• notebooks/ – data exploration & analysis in Jupyter
• src/ – production code organized by concern
– data/ – loading & preprocessing logic
– features/ – temperature compensation & feature engineering
– models/ – ensemble training & inference code
– visualization/ – plotting utilities for dashboards
– api/ – FastAPI app exposing prediction endpoints
• Config/build files: requirements.txt, README.md, INSTALL.md
• Tests directory (implied, though not in file_tree): tests/
Extract architecture insights from README
• Core ML pipeline stages: reading → cleaning/validation → feature engineering → model training/inference
• Ensemble of three models
• REST API with three POST endpoints
• Notebook-driven exploratory analysis and plotting
Detail without overthinking
• How modules call one another (e.g., API → src/data → src/features → src/models).
• Where tests and notebooks fit in: tests validate each module; notebooks provide exploratory insights.
• Document any configuration or dependency flows if space permits.