Nine layers.
One stack.
Written down.
The architecture, decision frameworks, and standards behind Orion's platform and AI work. Organized by the same nine layers as the Quantum Leap initiative — every engagement runs against this playbook.
Reference architecture, layer by layer.
Click any layer for the deep page — reference architecture, tools, build-vs-buy, and the five mistakes we keep seeing. The chips show where Orion does the substantive work and where we configure or evaluate bought primitives instead.
Bedrock-on-VPC, account isolation, KMS keys you own. Sprintsail primitives + raw CDK for AI-specific resources.
Source ingestion. Lineage. Document parsing. Permissioned at the row, with content-addressable hashes.
Claude on Bedrock by default. Per-task selection — Sonnet for reasoning, Haiku for routing. Versioned.
Hybrid dense + BM25 + RRF fusion. Layout-aware chunking. Permission filtering at the index, not after.
Single-turn pipelines by default. Agents only when the workflow earns the multi-step surface area.
Named owner per tool. Audit log with the prompt that triggered each call. Side effects behind human review.
Faithfulness, coverage, refusal correctness. The harness is the contract. Re-runs on every model bump.
Token spend, refusal rate, retrieval hit rate, tool-call success. One dashboard for finance and engineering.
Layered prompt-injection defense. PII masking before the prompt is built. Lineage in every audit record.
Tooling, and the vocabulary we use to talk about it.
Recent posts.
A Production Document-Intelligence Stack: The Nine Layers, End to End
What goes into a document-intelligence system that actually ships — layer by layer, with the choices that turn a demo into a production pipeline.
Read →Build vs. Buy at Every Layer of the AI Stack — A Framework
A nine-layer decision framework for what to build, what to buy, and what to defer when standing up a production AI system.
Read →Want depth on something specific?
We write the answer once and put it here. Send the question and a paragraph of context — happy to share early drafts on layers still in progress.