Platform · Norma

The explainability layer for regulated AI.

Norma sits between your AI models and your users, turning every recommendation into a decision card — a plain-language rationale, the evidence behind it, a confidence score, risk flags, and a recommended next action — with an immutable audit trail mapped to EU AI Act, CFPB, EEOC, and NIST.

Explainable decision card

Applicant #4821 — credit decision

87%

Decline explained: debt-to-income ratio of 48% exceeds the 43% policy threshold.

DTI 48% vs 43% maximum
Two credit inquiries in the last 90 days
Credit policy v3.1 applied
Watch: bureau score is 11 points from the approve band.
Issue a CFPB-compliant adverse action notice with the specific reasons.
How it works

One API between your model and your users.

01

Send a model output

POST the recommendation and its context to the /explain endpoint.

02

Get a decision card

Receive a structured, human-readable card your product renders anywhere.

03

Render anywhere

Drop-in React components or a plain JSON payload for any frontend.

04

Audit, always

Every card writes an immutable, exportable audit record mapped to the frameworks that apply to you.

Capabilities

From output to explanation in one layer.

Explain the recommendation

Convert opaque AI outputs into plain-language explanations a non-technical user can read — why the decision was made, what signals mattered, and how confident the system is.

Show the evidence trail

Surface the source records, documents, transcripts, or policies behind each recommendation so users can verify before they act.

Confidence & risk flags

A 0–100% confidence score plus an explicit list of what could be wrong — surfaced up front, not buried in a model card.

Audit export

Log the full reasoning trail and export it for reviews, approvals, and governance. Build your compliance record at the moment of the decision, not under audit.

Quickstart

Explain a decision in one call.

curl -X POST https://api.northgale.io/v1/explain \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model_output": "Applicant #4821 declined for a credit product.",
    "context": {
      "debt_to_income": 0.48,
      "policy_threshold": 0.43,
      "recent_inquiries": 2
    }
  }'

Returns a structured decision card — rationale, evidence, confidence, risk, and next action. Full reference in the docs.

Mapped to the obligations that apply to you.

Norma's audit trail maps to EU AI Act, CFPB adverse-action rules, EEOC, and NIST AI RMF — so your explanation is also your compliance evidence.

See the regulatory map →