How LargentIQ handles your applicants' data, how the AI is constrained, and what we'll show your auditor when they ask.
Every loan application flows through a deterministic, observable pipeline of five specialized agents. Each agent runs sequentially against the same application context; outputs are persisted as structured JSON before the next agent starts.
A human officer is always in the loop: agents recommend, officers decide. Every approval and rejection is logged with the officer's identity, timestamp, and optional note.
Each agent receives a strict system prompt with an output schema. A deterministic post-filter scrubs the customer-facing "Draft applicant message" output before it ever reaches the applicant — internal terms (KYC, AML, DTI, risk_score, etc.) are replaced with plain-language synonyms.
Every officer decision is recorded in audit_logs: officer id, officer email, application id, action (approved / rejected), bulk flag, timestamp, and the officer's free-text note. Bulk actions use a per-call sentinel UUID so only the apps actually modified by that call are logged — protecting prior decisions from being re-attributed.
On every approval, LargentIQ posts JSON to your configured webhook URL with an HMAC-SHA256 signature. Verify in any language:
Shared secret and webhook URL are configured per-tenant in the officer Settings page. Rotate either at any time without downtime.
What we have today, what we're building, and what we've committed to.
You can run your own status page at a domain your prospects recognize, e.g. status.yourbank.com — same page, same numbers, your branding.
status.yourbank.com, trust.yourbank.com, or sla.yourbank.com.CNAME status.yourbank.com → lending-decision-ai.emergent.host (TTL 300)./status-page renders on the new domain. Typical turnaround: 1 business day.Made with Emergent