Project semantic image search vision embeddings vector database participated in a hackathon focused on animal welfare. designed and implemented a photo - based matching system that identifies lost cats by comparing incoming photos against a database of registered animals using visual embeddings. used openai clip to encode cat photos into dense visual embeddings, capturing appearance, color, and pattern features without labeled training data. stored and queried embeddings using chromadb for fast approximate nearest - neighbor similarity search over the registered pet database. orchestrated the ingestion and matching pipeline with zenml, ensuring reproducible runs and clean artifact tracking. applied sklearn utilities for threshold tuning and similarity score calibration to control match confidence levels. stack: clip chromadb zenml sklearn

Customer churn prediction end - to - end ml system production ml pipeline ensemble models rest api interactive ui built a full production - grade churn prediction system from feature engineering and model training to a live api and user - facing dashboard. engineered with reproducibility and deployment in mind from day one. designed a full zenml orchestration pipeline covering data ingestion, preprocessing, training, evaluation, and model registration with artifact versioning. trained and tuned an ensemble of sklearn models (xgboost, lightgbm, randomforest) with optuna hyperparameter optimization, achieving roc - auc 0.87 on held - out test data. exposed predictions through a fastapi rest service with pydantic - validated request/ response schemas, health - check endpoint, and async inference. containerized the full stack with docker (multi - stage build for the api image) and wired together via docker compose. built a streamlit dashboard for real - time single - customer and batch prediction, with shap - based feature importance visualization. stack: zenml xgboost lightgbm sklearn optuna fastapi docker streamlit shap demo video: https: //www.youtube.com/watch v okevzeoar98 live api: https: //vester - project - api.onrender.com/docs live ui: https: //vester - ui.onrender.com note i use for hosting render free vesrion, so it will take a few minutes for loading