AI Master's Student @ FAU · IT Specialist @ TUM
Learning MLOps by building end to end: model training, experiment tracking, deployment, and monitoring.
ML: XGBoost · PyTorch · scikit-learn · MLflow · Evidently Serving: FastAPI · Docker · Docker Compose Orchestration: Apache Airflow · Kubernetes Monitoring: Prometheus · Grafana LLM / RAG: LangChain · ChromaDB · Anthropic Claude CI/CD: GitHub Actions · GHCR
| Project | What it does | Stack |
|---|---|---|
| churn-prediction · live demo | Churn API with model registry, drift + Prometheus monitoring, and Kubernetes manifests | XGBoost, MLflow, Evidently, Prometheus, Docker, K8s |
| price-prediction | California house-price regression API | XGBoost, MLflow, Evidently, FastAPI, Docker |
| anomaly-detection | Unsupervised sensor anomaly detection | IsolationForest, MLflow, Evidently, FastAPI, Docker |
| airflow-ml-pipeline | Scheduled training pipeline with drift-triggered retraining | Airflow, MLflow, Docker Compose |
| image-classifier | CIFAR-10 CNN served over an API | PyTorch, MLflow, FastAPI, Docker |
| document-qa | RAG document Q&A with source citations and retrieval evaluation | LangChain, ChromaDB, Claude, FastAPI, Docker |
| Project | What it does | Stack |
|---|---|---|
| gpu-scheduler | Fair-share GPU cluster scheduler with EASY backfill and starvation-free job aging | Python, FastAPI, Prometheus, Grafana, Docker |
| llm-inference-server | Continuous-batching inference server with iteration-level scheduling and TTFT/throughput metrics | PyTorch, Transformers, FastAPI, Prometheus, Docker |
Every project ships with tests, a Dockerfile, and a GitHub Actions pipeline.