Project Overview
The AI GitHub Developer Analyzer is a full-stack, portfolio-grade application designed to help recruiters and engineering managers make data-driven hiring decisions. By simply entering a GitHub username, the system fetches all public repository data via the GitHub REST API and runs it through a machine learning pipeline — delivering an instant, objective assessment of a developer's experience level.
The backend is powered by a FastAPI server that orchestrates a multi-stage data pipeline: extracting raw repository metrics (stars, forks, followers, language diversity), computing derived features like follower ratios and documentation quality scores, and feeding the final feature vector into a pre-trained Random Forest classification model for real-time inference.
The frontend is a high-performance, pure Vanilla HTML/CSS/JS application styled with a modern glassmorphism dark theme. It visualizes the prediction results on a dynamic dashboard — including a circular hireability gauge, a Chart.js doughnut chart for technology stack breakdown, and actionable AI-generated improvement suggestions tailored to the developer's specific profile metrics.
Key Features
- Multi-variate LSTM with attention mechanism for 72-hour ahead forecasting
- Integration of 8 pollutant features (PM2.5, PM10, NO2, SO2, O3, CO, temperature, humidity)
- Automated data ingestion pipeline from OpenAQ API (real-time sensor data)
- Sliding window dataset construction with configurable look-back window
- Comparison baseline: ARIMA, SARIMA, Prophet, Linear Regression
- Interactive Plotly dashboard with city-level forecast visualisations
- Health advisory alert system based on WHO AQI thresholds
- Uncertainty quantification via Monte Carlo Dropout inference
Technical Approach
The feature engineering pipeline extracts 9 key numerical features from the GitHub API response: public_repos, followers, following, total_stars, average_stars, total_forks, tech_diversity_score, documentation_score, and a derived follower_ratio. These features are carefully selected to capture both the quantity and quality of a developer's open-source presence.
The classification model is a Random Forest Classifier trained on 2,000 synthetically generated developer profiles using sklearn. The synthetic dataset was constructed with statistically realistic feature distributions for each seniority tier, using numpy with a fixed random seed (42) to ensure reproducibility. The model was trained on an 80/20 train-test split and achieves high accuracy for standard GitHub profiles.
The hireability score is computed by a custom weighted scoring engine that runs independently of the ML model. It distributes 100 points across four categories: Repository Popularity (40pts — stars & forks), Activity & Volume (30pts — repo count & followers), Technology Diversity (15pts), and Documentation Quality (15pts). Caps are applied at each bucket to prevent viral outlier repositories from skewing the final score.