Prompt Laboratory makes prompts reviewable software assets: YAML contracts define the interface, repeatable datasets exercise behaviour, evaluators create evidence, and regression policies support human release decisions.
PROBLEM
Why it exists
Prompts often become production dependencies while remaining unversioned strings with weak test evidence, inconsistent review and provider-specific behaviour.
ARCHITECTURE
System flow
Figure 1 · Git-controlled prompt lifecycle from contract and dataset to evidence-backed approval.
CAPABILITIES
What is implemented
▸ YAML prompt contracts with versions, owners, variables and output schemas
▸ Strict Jinja2 rendering with Pydantic input validation
▸ Cross-industry synthetic evaluation datasets
▸ Deterministic checks plus optional LLM-as-judge evaluation
▸ OpenAI, Anthropic, Gemini, local-compatible and mock providers
▸ Side-by-side quality, latency, token and estimated-cost comparison
▸ FastAPI, Streamlit and PostgreSQL product surfaces
▸ Authentication, workspaces and encrypted per-user provider credentials
