My projects
Click on a project to access its GitHub repository.
Containerized ML Inference API (FastAPI + Docker + MLflow)
FastAPI inference API packaged with Docker, featuring a pre-trained RandomForest (Iris). Experiment tracking with MLflow (UI via docker-compose). Endpoints: /predict, /predict_batch, and /health. Training script logs params/metrics and regenerates models/model.pkl. Reproducible deployment and testing with Docker.
ML Experiments Tracking with MLflow (mlflow)
Full MLflow demo: experiment tracking (params, metrics, artifacts), run comparison via UI, autologging, and model registry/versioning. Trained across multiple datasets (Wine, Iris, Breast Cancer, Digits, California Housing) with various algorithms (Logistic Regression, RandomForest, SVC, KNN, GradientBoosting), scaling pipelines for convergence, logged GridSearchCV, and local REST model serving.
Credit Card Fraud Detection (fraud-detection)
End‑to‑end fraud detection on a Kaggle dataset. EDA, class imbalance handling (SMOTE, class_weight), model training (RandomForest, Logistic Regression), decision threshold tuning (F1, recall), ROC/PR analysis, and feature importance. Deployed a Streamlit app to test the model in realistic scenarios.
Plots cumulative returns and computes performance indicators
Streamlit app comparing basics strategies like Buy & Hold, SMA50, RSI, and Donchian on Yahoo Finance data.
AI Game
Complete Tic‑Tac‑Toe in Python with Tkinter GUI. PvP and vs Bot with three difficulty levels (easy, medium, hard via minimax). Includes smooth animations, light/dark theme, scoreboard, hover highlights, win highlights, and visual effects (confetti).
Macros and Markets
Live Market plus tools: EURUSD/SPY/QQQ/ES/NQ quotes and key rates (Fed/ECB, US/DE 10Y). P/E & beta for SPY/QQQ and mega-cap tech (NVDA, AMD, MSFT, AAPL, AMZN, META, NFLX), and a configurable cross-asset correlation heatmap.
MLOps Finance — Portfolio Forecast & Backtest Pipeline
End-to-end pipeline: ingestion (yfinance) → features (returns/vol/RSI/MACD/lags) → models (Ridge, Logistic, XGBoost, ARIMA) → backtests (threshold + walk-forward) → reporting (HTML) → API (FastAPI) → orchestration (Airflow + MLflow). Includes drift monitoring and file pruning.
Heart Disease Prediction — Risk Factors & ML Models
Predictive modeling project using the UCI Heart Disease dataset: risk factor analysis (age, sex, cholesterol, ECG, etc.) → models (Logistic Regression, Random Forest, XGBoost) → evaluation (cross-validation, hyperparameter tuning, ROC-AUC ≈ 0.96) → medical insights (key features: thal, ca, oldpeak, cp, thalach). Includes pipeline with imputation, scaling, and clear reporting.