Reference architecture · Strata
The architecture for turning scattered operations into governed systems.
Strata is the reusable architecture underneath our engagements: ingestion, entity resolution, event detection, decision rules, evidence and user-facing experience. It helps us move faster without asking you to buy a platform or replace the systems you already run.
01
The layers
You cannot make the top layer trustworthy by ignoring the layers below it.
Dashboards, AI assistants and workflow tools are the visible layer. They become useful only when the data, entities, events and rules underneath are governed.
Experience
Query, reporting and the interfaces people actually use. Built last, deliberately — an interface over an ungoverned model is a faster way to be wrong.
Decisioning
Policy-driven decisions expressed against the model, not buried in application code where nobody can audit them.
Event detection
What changed, when, and whether it matters. The difference between a dashboard you check and a system that tells you.
Entity resolution
Deciding that this dealer record and that one are the same dealer. Unglamorous, and the single largest source of contradictory numbers we’ve found in any organisation.
Ingestion
Everything arriving in whatever shape it arrives in — feeds, exports, spreadsheets, inboxes — captured without demanding the sender change first.
02
Built from real components
Mainstream infrastructure, by design.
Kafka or equivalent event backbones, Postgres before an exotic database, Neo4j where relationships matter more than tables, Qdrant for retrieval that has to be grounded rather than guessed. Deliberately familiar — the differentiated value sits in the model and the contracts, not in infrastructure you’d have to hire specialists to maintain after we leave.
Evidence-first
Every derived fact traces back to its source. No confident claim without a receipt.
Replay-safe
Reprocessing creates new versioned records. Nothing gets silently overwritten.
Policy-governed
Promotion, review and decisioning are inspectable, not buried in one engineer’s judgment.
Multi-tenant by default
Built to serve hundreds of tenants correctly the first time, not retrofitted after tenant 40 breaks it.
03
Proven, not pitched
Every reusable pattern came from production work.
RYNO’s RAG layer runs on a Qdrant vector store. Scalekrut’s matching engine runs on a Neo4j graph. RYNO’s ingestion runs on Airflow through Glue/Spark into an S3 lake. None of it was built speculatively — each piece earned its place in Strata by solving a real client’s problem before it was named as a reusable pattern.
04
Strata’s first commercial product
Strata Exception Intelligence.
Automatically investigate operational exceptions, explain their cause using evidence, recommend the next action, and optionally execute it under policy. The same five layers above, packaged around one problem: something broke, and someone needs to know why before they can fix it.
See Exception Intelligence in full05
How it’s packaged
Scoped to the problem, not sold as a platform licence.
Four tiers, described by what’s covered rather than a fixed price — final scope and commercials get set per conversation, the same way every engagement does.
Starter
One operational workflow, up to 3–4 source systems, the evidence and context layer, exception investigation, human review. Built to prove value on one real problem before anything expands.
Professional
2–3 workflows, 5–8 source systems, entity resolution, event intelligence, policy-based decisions and recommendations, governed actions.
Operations
Multiple operational domains, a larger source footprint, an advanced entity graph, process intelligence, action orchestration, replay and quality monitoring.
Enterprise
Enterprise-wide operational intelligence, priced against domain count, data volume, environments and compliance requirements. Negotiated, not listed.
Every tier keeps the same discipline as the full engagement: open stack, documented model, evidence before conclusions.
Next