Watch the substrate run
A working simulation of what happens underneath a BasinWright deployment: records arriving unresolved, a hub deciding what is fit to use, three engines grounding and verifying a case, and a decision that keeps the lineage of every record behind it. Configure it around your own systems and it runs your board instead of ours.
We build it, run it and watch it. You own it.
The seven stages above, written out. Each chapter lights the part of the substrate it describes, so what you just watched a case do is what you are about to read.
- 01Ingest
Everything arrives messy
Records land unnormalised, duplicated and unlinked, from systems that never agreed on a schema. They enter the substrate grey, because nothing has earned meaning yet. Colour here is not decoration — it is the quality of the record.
UnnormalisedDuplicatedUnlinkedUnknown quality - 02Cognitive Data Hub
Resolution is where meaning starts
The hub resolves entities, collapses duplicates into one governed record and decides what is fit to use. What fails the quality bar falls into quarantine — some records are remediated and released, others expire. A funnel with no rejects is not one anybody has run.
Entity resolutionDuplicate collapseQuarantine & remediationPolicy applied once - 03Build & deploy
BasinWright builds the model and deploys it into your estate
We train, tune and evaluate against your governed data, then deploy into your tenancy and your region — not ours. The control plane underneath the field is our work; the estate it runs in is yours.
Trained on your dataEvaluated before promotionDeployed in your tenancyYour region, your keys - 04Ground · Verify · Explain
Three engines, running at once
Cognitive RAG grounds the case in retrieved evidence, deterministic models verify it against real constraints, and LLM reasoning weighs the options and explains the call. Three concurrent lanes, not three sequential steps.
Cognitive RAGDeterministic modelsLLM reasoning - 05Run & monitor 24×7
Monitored every hour of every day
Drift, accuracy, latency and cost are watched continuously, and a model that starts to slip is retrained and re-evaluated before it reaches a decision that matters. Running it is the part that never stops — and it is the part we carry.
Drift detectionAccuracy & latency SLOsCost per decisionRetrain and roll forward - 06Decision layer
A case decides only when every lane is satisfied
Evidence accrues per lane and nothing is decided early. The agent council then runs its skills over the assembled evidence, and every decision keeps the lineage of the exact records that produced it. Confidence is computed, not asserted.
Evidence quorumAgent councilComputed confidenceRecord-level lineage - 07Ownership
Your business owns the model and the intelligence
We build it, deploy it and keep it running — but what comes out the other end is yours. The model, the weights, the governed data it learned from and every decision it has ever made stay inside your estate, under your control. Intelligence you own and can walk away with, not a black box you rent by the month.
You own the modelYou own the dataYou own the decisionsNo lock-in
Run it against your own systems
The console you just configured travels with your enquiry — the sources you connected and the use cases you picked arrive with it, so the first conversation starts from your estate rather than from ours.