P-01
MCP Gateway
A single layer that exposes sources and tools via MCP, with centralized policies, auditing and caching.
View pattern →P-02
Permission-aware RAG
Retrieval filters by who is asking before ranking, so the model never sees what the user is not allowed to see.
View pattern →P-03
Budgeted agent with human in the loop
The agent runs within limits on steps, cost and permissions; irreversible actions go through approval.
View pattern →P-04
Continuous evaluation loop
Real production failures become test cases that every new prompt or model version must pass.
View pattern →P-05
End-to-end LLM tracing
One identifier ties together question, retrieved context, prompt, model, tools, answer and feedback.
View pattern →P-06
AI over legacy with an anti-corruption layer
AI talks to a clean domain model; a layer translates and protects the legacy system.
View pattern →P-07
Cost-aware routing with caching
Each task goes to the cheapest path that meets the required quality, with caching first.
View pattern →