OPEN-SOURCE · SELF-HOSTED · HANDS-ON
Build an inspectable LLM quality loop.
Observe every call, evaluate changes, and query the evidence in ClickHouse.
- 2–3.5h
- hands-on path
- 10
- guided modules
- 4
- deployment paths
LLMOps in a Box¶
Use the top navigation to move between the learner workshop and operational references. Inside the Workshop tab, the left navigation keeps the complete module sequence visible from setup through ClickHouse analytics and wrap-up.
Start here¶
New participants should begin with 00 — Setup and LLM Connection. Instructors can open the Instructor Guide, while maintainers can use the Docker Validation Report before a customer delivery.
What you will build¶
model request → observable trace → managed prompt → golden dataset
→ repeatable experiment → automated and human scores
→ ClickHouse quality analysis → ship/no-ship decision
The workshop supports four explicit deployment paths without provider fallback: ClickHouse Cloud or local ClickHouse, each paired with Anthropic Sonnet or a participant-selected local Ollama model.