End-to-end ML engineering — from data ingestion and feature engineering through model training, experiment tracking, and serving. Most of these came out of real fintech datasets and production constraints.
Rock Paper Scissors played with real hand gestures via webcam. Uses MediaPipe for real-time hand landmark detection and an adaptive Markov chain CPU opponent that learns from your play patterns.
Graph-based image segmentation implemented from scratch. Models an image as a weighted undirected graph — each pixel a node, edges weighted by color similarity. Union-Find (with path compression + union-by-size) merges the most similar pixels first. Tunable k, σ, and min_size parameters give coarse-to-fine control over segmentation granularity.
End-to-end ML pipeline for credit scoring on real fintech loan data. Feature engineering, model selection (XGBoost, LightGBM), MLflow experiment tracking, and a FastAPI inference endpoint.
Churn prediction system with class-imbalance handling (SMOTE), SHAP explainability, and a Dagster orchestration pipeline pulling from PostgreSQL via dbt-transformed tables.
Real-time fraud detection using transaction sequence features, geospatial signals, and an ensemble model. Includes a streaming-friendly inference layer and anomaly scoring dashboard.
Multi-horizon demand forecasting pipeline over financial telemetry data. Compared ARIMA, Prophet, and LSTM approaches with automated retraining on drift detection.
Text classification and entity extraction pipeline for financial documents. Fine-tuned a transformer model for Amharic-English mixed text and deployed with a FastAPI wrapper.