AutoMap
Type inference and field-resolution intelligence. CleanAI inspects column headers, sample values, and downstream taxonomies to propose deterministic mappings — confidence-scored, steward-reviewed, never auto-deployed.
CLEANAI · INTELLIGENCE LAYERThe intelligence that
The intelligence that
never executes alone.
THE INTELLIGENCE LAYER
CleanAI is the reasoning engine inside CleanFlowAI. It drafts rules, infers types, reconciles schemas, and surfaces rationale for every quarantined record — but it never writes to your data. Every output is a proposal, every proposal carries evidence, every approved artefact becomes a registered CleanDataShield template.
THE REASONING SURFACES
Four reasoning surfaces, one engine.
CleanAI is composed of four intelligence surfaces — each tuned for a discrete data-trust task, each governed by the same approve-before-execute contract.
Business Rules Suggestion
Plain-English rule definitions compiled into deterministic CleanDataShield templates. CleanAI drafts; your stewards review and approve; CleanDataShield enforces. No arbitrary code reaches production.
Reasoning for Quarantine
Every quarantined record arrives in the editor with CleanAI's rationale: which rule fired, what was malformed, what remediation it recommends. Reviewers approve, override, or escalate — nothing executes without human sign-off.
AI-Assisted Support
CleanAI surfaces remediation playbooks, blueprint templates, and rule-pack suggestions in-context — grounded in your registered taxonomies, never inventing operations the platform doesn't support.
THE APPROVE-BEFORE-EXECUTE LOOP
Stages between a draft and a deployed rule.
- STAGE · 01
Suggest
CleanAI proposes — type inferences, rule drafts, mapping candidates, remediation recommendations. Every output ships with confidence scoring and evidence.
- STAGE · 02
Approve
Your stewards review every proposal in context. Approve, override, refine, or reject. Nothing enters production without explicit sign-off.
- STAGE · 03
Register
Approved artefacts become first-class platform objects — versioned templates, named blueprints, registered Jobs. Never one-off scripts.
- STAGE · 04
Execute
CleanDataShield enforces only registered templates. Every run is rule-attributed, version-pointed, and immutably logged.
What makes CleanAI audit-ready.
AI suggests. Humans approve.
CleanAI is an intelligence layer, not an execution layer. Every proposal — type inference, rule draft, mapping suggestion, remediation recommendation — requires explicit steward approval before it touches data.
Deterministic execution only.
No model output runs against production data. CleanAI's drafts compile into registered, version-controlled CleanDataShield templates that execute deterministically and identically every run.
Evidence with every output.
Every CleanAI output carries its provenance: input snapshot, confidence score, rule lineage, and steward attribution. Reviewers see the reasoning, not just the verdict.
Bounded by your taxonomy.
CleanAI operates against your registered taxonomies, schemas, and rule library. It cannot invent operations the platform doesn't support, and it cannot escape your governance perimeter.
Four principles.
One philosophy.
One pipeline. Thirty-four rules. Every record accounted for — validated, remediated, or quarantined.
Ingest a CSV. Retrieve a DQ report. No ETL glue, no custom scripts, no surprises.
AI suggests. Human approves. Pipeline executes deterministically — always.
If we can't remediate it, we quarantine it. If we quarantine it, we surface the rationale.
0arbitrary code paths in production
“Our security team cleared CleanFlowAI in one session. CleanAI drafts, CleanDataShield enforces, and every line is reviewed by a steward. There was no attack surface to debate.
CLEANAI ACROSS THE PLATFORM