Practical Data Science

End-to-end data science on a real on-prem GPU server. Stand up a multi-user ML platform — JupyterHub, MLflow, MinIO, Slurm, Prefect, dbt, vLLM, Label Studio, Prometheus — then use it to ship four capstones: tabular ML, a modeled warehouse with dashboards, a self-hosted RAG application, and a multi-GPU computer-vision model. 2× NVIDIA L40S included.

17 modules · ~27h total reading time

Start the track →
Show all modules 17
00 — Overview
01 — Infrastructure: the GPU server
02 — The toolchain
03 — Reproducibility stack
04 — GPU scheduling with Slurm
05 — The data lake
06 — The warehouse
07 — Orchestration with Prefect
08 — Superset (Capstone 2)
09 — Tabular ML (Capstone 1)
10 — Serving and drift
11 — Self-hosted LLM (vLLM)
12 — RAG (Capstone 3)
13 — LoRA fine-tuning
14 — Computer vision (Capstone 4)
15 — CI/CD with Gitea Actions
16 — What's next