Governed AI operations
for mortgage lenders.
Miji runs prequalification end to end — application intake, document verification, credit, and calculations — with human approval gates and a tamper-evident record of every action.
Now onboarding lending design partners. A product of Canopy Mind, Inc.
Operations
Prequalification
Total applications
128
+12% vs last 30 days
Approval rate
64%
awaiting decisions: 9
Pending / incomplete
17
in progress or awaiting approval
Avg time to decision
3.2m
last 30 days
every run audited · approvals routed to your reviewers
What Miji runs for lenders right now.
Not a roadmap — the working product. These capabilities run mortgage prequalification operations today.
URLA-1003 intake
Native application forms with versioning, completion tracking, and PDF import (native + OCR + AI extraction).
Document processing
Extraction with per-field provenance — every candidate value traces to its source document, page, and confidence.
Verification policies
Policy-controlled auto-verification: minimum confidence, agreeing sources, tolerances, and human-required fields.
Credit integration
Equifax credit report pulls inside the governed operation, with access logged like every other action.
Deterministic calculations
DTI and qualification math run on the workflow engine — repeatable, auditable, replayable. Not an LLM's arithmetic.
Human approval gates
Threshold-based routing to approval groups — any-one, all, or majority strategies — with full decision records.
Review workbench
Exception-based case review: your team sees blocking findings, conflicts, and low-confidence fields — not every file.
Tamper-evident audit
Hash-chained audit trail with continuous chain verification. Every action, decision, and approval — provable.
LOS submission
Completed, verified applications hand off to your loan origination system. Miji augments your stack — it doesn't replace it.
The GSEs now require AI governance. From every lender.
Freddie Mac Bulletin 2025-16 took effect March 3, 2026. Fannie Mae Lender Letter LL-2026-04 takes effect August 6, 2026. Selling to the GSEs now means demonstrating governance over the AI in your operation — including AI you buy.
Written AI governance
Documented policies covering how AI is used, monitored, and controlled across your origination process.
Bias monitoring & auditability
The ability to monitor AI-influenced outcomes and produce audit records when the GSEs or examiners ask.
Accountable senior officer
A named executive responsible for AI use — with the visibility to actually stand behind that accountability.
Vendor AI included
The obligations explicitly extend to AI systems your vendors provide — buying AI doesn't outsource the duty.
Most AI vendors answer this with a security certification. It doesn't answer it. What answers it is a per-loan record of every AI action, every approval, and every decision — with integrity you can prove. That is what Miji is built to produce as a by-product of running the work.
Miji is not affiliated with or endorsed by Fannie Mae or Freddie Mac. This summary is general information, not legal or compliance advice — consult the source documents and your counsel.
The GSE Compliance Pack.
Built on Miji's audit chain and decision records — shaped with our design partners so it matches how your exams actually run. Partners get early access and direct influence on what ships first.
Per-loan AI usage register
An automatically generated inventory of every AI action in a loan's lifecycle: what ran, what data it touched, what it produced, and whether a human reviewed it.
Examiner export
A hash-verified bundle per loan or exam period — decision records, approvals, overrides — with a chain-verification report the examiner can run themselves.
Governance policy templates
The written AI/ML governance policy the mandates require, pre-mapped to how Miji's controls satisfy each clause.
Officer attestation view
A periodic dashboard for the accountable senior officer: what AI ran, at what autonomy, exception and override rates, chain-integrity status — with sign-off.
Fair-lending data exports
Structured decision-outcome exports designed to feed your fair-lending analysis — outcomes, reasons fired, and overrides, queryable by period.
Why this can't be bolted on
These artifacts are only as trustworthy as the record beneath them. Miji's hash-chained audit trail and deterministic replay are already built — the pack is the reporting layer on top, not a retrofit of ordinary logs.
Verification policy
Active v1 · 2 version(s)
Minimum confidence
0.92
Minimum agreeing sources
2
Numeric tolerance
1.5%
Date tolerance
3 days
Published versions are immutable; activating a version never alters existing verified facts.
Gross Monthly Income
criticalrequiredVerified Fact (v2)
$8,450.00 · human verified
Your credit policy, enforced.
Verification policies turn your overlays into executable rules: which facts must be verified, from which sources, with what confidence and tolerances — and exactly when a human must review.
Versioned and immutable
Policies publish as immutable versions. Which policy governed which loan always has an exact answer — the question examiners actually ask.
Consistent by construction
Identical files get identical treatment. Policy-enforced execution removes processor-to-processor variance — a fair-lending asset, not just an efficiency.
Facts, not vibes
Every verified fact records its sources, confidence, method, and reason code. Historical facts are never deleted — corrections supersede, with full history.
On the roadmap: policy simulation
Replay a proposed policy version against past loans before activating it — see what would have changed, before it changes. In development with design partners.
Between your POS and your LOS — not instead of them.
Miji is the governed operations layer. Leads and applications come in from the systems you already run; verified, decision-ready files hand off to your LOS. No rip-and-replace.
Intake
Your POS / CRM / leads
Wherever borrowers arrive today
Miji
Governed AI operations
Document verification · credit · calculations · approval gates · tamper-evident audit
Origination
Your LOS
Verified, decision-ready files
Vendor-neutral by design: Miji connects to your existing systems with scoped, credentialed access — and because it's model-agnostic, your AI operation isn't hostage to any single model vendor.
Cost discipline is the differentiator now.
In a stable-rate market, volume won't save the P&L — operational efficiency will. Personnel is roughly two-thirds of the cost to originate, and prequalification is where leads are won or lost.
$11,898
Cost to originate one loan
IMB production expense, Q1 2026
$727
Net production profit per loan
What's left after that cost
$2,529
Per-loan gap, best vs. worst operators
Top vs. bottom cost quintile
Miji attacks both sides: speed to lead — instant, governed prequalification while the borrower is still on the phone — and exception-based processing, where your team reviews what needs judgment instead of touching every file. Every hour of AI work is metered, budgeted, and attributed per operation.
Cost figures: Mortgage Bankers Association quarterly performance reports (Q1 2026) and MBA Chart of the Week, April 2026.
The mandates are here. Build your answer with us.
We're onboarding a small group of lending design partners — IMBs and mid-size lenders who want governed AI operations they can defend in an exam.
- Early access to the GSE Compliance Pack — and direct influence on what ships first
- A governed prequalification operation configured to your credit policy
- Self-hosted deployment option for banks and lenders with data-residency requirements
- Design-partner pricing, locked in
hello@mijios.ai · lender.mijios.ai