How ykato27/careernavigator works
Main components
a. Data Stores
• Input CSV directories under data/ (members, acquired, skills, education, license, categories)
• Output CSV directory output/
b. Core Library (skillnote_recommendation package)
1. Data Ingestion & Transformation
• config_loader.py / core/config.py
• core/data_loader.py
• core/data_transformer.py
2. Rule‐Based Recommendation
• core/similarity_calculator.py
• core/recommendation_engine.py
• core/recommendation_system.py
• core/evaluator.py
3. Machine‐Learning Recommendation
• ml/matrix_factorization.py
• ml/ml_recommender.py
• ml/diversity.py
4. Knowledge‐Graph & Visualization
• graph/knowledge_graph.py
• graph/hybrid_recommender.py
• graph/random_walk.py
• graph/career_path.py + career_path_visualizer.py
• graph/path_visualizer.py + visualization_utils.py
5. Utility Modules
• utils/streamlit_helpers.py
• utils/visualization.py
c. Execution Interfaces
• Streamlit Web UI (streamlit_app.py + pages/1_Model_Training.py, pages/2_Inference.py)
• CLI Scripts (scripts/convert_data.py, scripts/run_recommendation.py)
d. Tests & Documentation
• tests/…
• docs/…
Relationships & data flows
• Input CSVs → DataLoader → DataTransformer → “Clean” DataFrames →
• Rule‐Based Engine → RecommendationSystem
• MLRecommender → (optional) Diversity module → RecommendationSystem
• RecommendationSystem → Evaluator (for offline metrics)
• RecommendationSystem & Graph modules → Web UI / CLI
• Web UI & CLI → Output CSVs / On‐screen visualizations
Architectural patterns & design principles
• Layered architecture:
• Data layer (CSV files)
• Processing layer (ingestion, transformation)
• Business logic layer (rule‐based + ML engines)
• Presentation layer (Streamlit UI + CLI)
• Modular design: clear separation of core modules (rule‐based) vs. ML modules vs. graph/visualization.
• Extensibility: configuration via YAML, pluggable RecommendationEngine subclassing, ML‐diversity strategies.
Technologies & frameworks
• Python 3.11+
• pandas / numpy (data handling)
• scikit‐learn & NMF (ML recommender)
• Streamlit (web frontend)
• networkx or custom graph code (knowledge graph)
• uv (environment + CLI)
• Clear component labels (e.g., “DataTransformer”, “MLRecommender”)
• Arrowheads for data flow direction