AI Architecture Discussions
A reference architecture for enterprise AI, laid out layer by layer. It is a way to organise the problem — grounding, orchestration, operations, and the guardrails around them — not a product we sell or a system in production. Start with the blueprint, then follow each layer into its own diagram and the components it contains.
The whole picture on one page: four layers read top to bottom — interface, the LLM platform, operations, and the data foundation — framed on both sides by the system-wide guardrails. Each layer below expands into its own diagram.
The layers
Interface & Consumption
3 componentsWhere the platform meets people: copilots inside the tools they already use, AI folded invisibly into productivity apps, and analytics driven by plain questions rather than a report queue.
Open →Enterprise LLM Platform
5 componentsThe central intelligence layer: the request lifecycle from prompt registry through orchestration to a multi-model hub, with serving and RAG grounding on the path.
Open →Operations & Integration
3 componentsThe runtime edge: a gateway that fronts the platform, agents that plan and act, and the connectors that reach the systems of record where transactions actually happen.
Open →Data Foundation
4 componentsHow context reaches the model: the source repositories, the ingestion pipeline that makes them searchable, the vector store, and the knowledge graph that maps the relationships between them.
Open →System-Wide Guardrails
6 componentsThe governance that wraps the rest: access control, cost, observability, audit, fairness, and evaluation — grouped into security, safety, and operational telemetry, and applied across every layer.
Open →