The Loan Origination System Built for AI — Not Retrofitted for It
298 routes from borrower application through funded. 9 specialized AI agents. Deterministic Fannie Mae 1084 income calculation. 180+ Encompass bidirectional fields. Cycle time reduced from a 45-day industry average (MBA Q4 2024) to under 4 days in the April 2026 production run.
Confer LOS is a cloud-native, full-stack mortgage loan origination system covering the complete loan lifecycle — from borrower application through funded — for mid-size mortgage lenders, banks, and credit unions. Unlike legacy platforms bolted with AI afterthoughts, Confer is AI-native: 9 specialized agents handle document classification, income calculation, condition generation, employment verification, borrower communications, and compliance monitoring as first-class system functions. The platform ships with 298 routes, a deterministic Fannie Mae 1084 income calculator (no LLM in the math path), TRID compliance timers backed by Temporal durable workflows, and bidirectional sync with 180+ Encompass fields — making it both a standalone replacement and a compatible complement to existing ICE Mortgage Technology infrastructure.
Core Capabilities
9 AI agents, each built for a specific function in the mortgage workflow.
Document Intelligence
3-tier AI classification (pattern matching → LLM → computer vision) with structured parsers for W-2, pay stub, 1040, bank statements, and driver's licenses. Automates the document processing workflow that typically takes 40–60 minutes per loan manually.
Income Calculation Engine
W-2, Schedule C, Schedule E, K-1, investment, retirement, and other income; fully deterministic. No LLM in the math path. 2-year trending, agency add-backs, and automated variance flags.
Underwriting Engine
Built-in AUS engine (~2,700 lines) with MISMO 3.4 export for DU/LPA submission. Configurable business rules engine, automated condition generation, and an AI underwriting agent with full audit trail.
Compliance Engine
TRID timers run inside Temporal durable workflows. HMDA auto-populates 110+ LAR fields. Covers QM/ATR (8 factors), HOEPA, ECOA, RESPA, and post-close QC with immutable audit logging.
Encompass Integration
OAuth-authenticated direct API access to ICE Mortgage Technology. All 15 E-Folder containers mapped. AI-classified documents filed automatically. Milestone sync and ULDD/MISMO secondary market exports.
Voice AI — Kylie
VAPI-powered voice assistant handles inbound loan status inquiries, outstanding condition review, and document submission prompts. Supports outbound reminder calls and transfers to loan officers.
Confer vs. Encompass — capability comparison
Feature and architecture comparison vs. leading mortgage LOS platforms
| Feature | Confer LOS | Encompass (ICE) | Blend | Blue Sage |
|---|---|---|---|---|
| AI architecture | AI-native — 9 agents in core workflows | AI via third-party plugins | POS-layer AI assist | Limited automation |
| Built-in AI agents | 9 production agents | None native | Partial | None |
| MCP tool protocol (open standard) | ✓ 32+ tools | ✗ | ✗ | ✗ |
| Deterministic 1084 income calc | ✓ 7 types, no LLM in math | External integrations required | Not included | Limited |
| Self-employed income (Sch. C/K-1) | ✓ Native, 1084-compliant | Requires third-party integration | ✗ | Manual |
| TRID durable timers | ✓ Temporal — survive restarts | Rule-based alerts | ✗ | Rule-based |
| HMDA auto-population | ✓ 110+ LAR fields | ✓ | Partial | Partial |
| Encompass sync | ✓ 180+ fields, bidirectional | Native | Via connector | Via connector |
| Voice AI | ✓ VAPI — inbound + outbound | ✗ | ✗ | ✗ |
| Document auto-classification | ✓ 3-tier (pattern + LLM + vision) | Third-party required | Partial | ✗ |
| Full lifecycle (app to servicing) | ✓ 298 routes | ✓ | POS only | Partial |
| Open architecture (any LLM) | ✓ MCP standard | Closed SDK | Closed | Closed |
| Real-time collaboration | ✓ Field-level locking, presence | ✗ | ✗ | ✗ |
Technical Specifications
Source: codebase analysis, March 2026
| Total pages / routes | 298 |
| API endpoints | 70+ |
| AI agents (production) | 9 specialized |
| MCP tools (total) | 32+ |
| Income type calculators | 7 (W-2, Sch. C, Sch. E, K-1, investment, retirement, other) |
| Encompass field mappings | 180+ (bidirectional) |
| E-Folder containers mapped | 15 (U1–U15) |
| Borrower application steps | 42-step progressive wizard |
| Staff portal roles | 7 (admin, LO, processor, underwriter, closer, correspondent, servicing) |
| Supported loan types | Conventional, FHA, VA, USDA, Jumbo |
| AUS engine | Built-in (~2,700 lines) + MISMO 3.4 export for DU/LPA |
| Temporal durable workflows | 8 (TRID timers, loan lifecycle, VOE, document pipeline, escalation) |
| Loan pipeline stages | 10 (Application through Funded/Denied/Withdrawn) |
| Communication channels | 5 (Email/Resend, SMS/Twilio, Voice/VAPI, WhatsApp, In-App) |
| ATR verification factors | 8 (per 12 CFR § 1026.43(c)) |
| HMDA LAR fields auto-populated | 110+ |
| AI classification confidence threshold | 90% |
| Front end | Next.js 16 App Router, React 19, TypeScript 5.7, Tailwind CSS 4 |
| Database | PostgreSQL via Supabase (Row-Level Security, multi-tenant) |
| Secondary market exports | ULDD, MISMO 3.4 |
Frequently Asked Questions
See a real loan run end-to-end in 30 minutes.
Walkthrough with the founding team. We'll bring the April 2026 production trace; you bring three of your typical files. We'll show you where Confer would change the cycle time on your actual workflow.
Deployed by independent mortgage banks, community banks, and cross-border lenders.