SmartAdvisor™ Technical Whitepaper

1. Abstract

SmartAdvisor™ is an AI-powered sidebar embedded in the MortgageTech® URLA Workbook — a custom tool for ICE Mortgage Technology's Encompass Web platform. It analyzes loan data in real time, surfaces actionable insights, tracks workflow progress, monitors conditions and documents, and provides conversational AI assistance — all within the originator's existing workflow.

This whitepaper describes the technical architecture, AI integration approach, and business value of SmartAdvisor™ for mortgage lenders seeking to reduce cycle times, improve data quality, and enhance compliance.

2. The Problem: Fragmented Loan Intelligence

Modern mortgage origination involves hundreds of data fields, dozens of conditions, complex regulatory timelines, and multiple stakeholders. Yet most LOS platforms present this information in scattered screens with no contextual intelligence.

The result: Loan officers catch DTI problems after submission. Processors miss expired disclosures. Underwriters re-request documents that were already in the eFolder. These inefficiencies add an average of 7-12 days to origination cycles and contribute to fallout rates of 20-30% industry-wide.

47d

Avg. Close Time

23%

Industry Fallout

831+

Fields per Loan

6

Roles per File

3. The Solution: SmartAdvisor™

SmartAdvisor™ brings real-time intelligence directly into the originator's workflow. Instead of switching between screens to check ratios, review conditions, or verify documents, everything appears in a single sidebar that updates as loan data changes.

✨

AI-Powered Insights

Azure OpenAI analyzes loan profile and returns categorized risk, compliance, and informational cards with field-level navigation.

📊

Real-Time Calculations

LTV, DTI, housing ratio, P&I — all computed client-side with zero round-trips. Updates as fields change.

🤖

AUS Simulation

Client-side DU/LP/Total Scorecard simulation. Pre-screen loans before submitting to the AUS.

📋

Conditions Tracker

23 condition types tracked with auto-matching to eFolder documents. Open/satisfied status in real time.

⏰

TRID Compliance

LE/CD timing, waiting periods, earliest close dates — all calculated with automatic alerts for expirations.

🔄

Role-Based Workflows

Six roles, each with tailored checklists, conditions views, and handoff tracking. No one sees irrelevant data.

🔍

Smart Navigation

Click a field ID in any insight card and SmartAdvisor™ opens the correct panel, scrolls to the field, and highlights it.

💬

Conversational AI

Ask questions about the loan in plain English. Grounded in actual loan data, not generic advice.

4. Technical Architecture

SmartAdvisor™ is built as a pure client-side JavaScript module (~5,200 lines) that loads inside the URLA Workbook. It has zero npm dependencies and runs in any modern browser.

4.1 Module Structure

// Core modules
smart-advisor-live.js    → Main IIFE module (5,200 lines)
smart-advisor-panel.html  → HTML template (268 lines)
smart-advisor.css        → Styles + dark mode (1,910 lines)
workflow-engine.js       → Calculation + AUS engine (3,267 lines)

// Supporting modules
encompassBridge.js       → SSF ↔ Workbook data bridge
encompass-navigator.js   → Field search + panel navigation
encompass-document-viewer.js → eFolder integration
fieldLockManager.js      → Field-level access control
FieldHelpTooltip.js      → 831+ field help definitions

// Server-side (Azure Function)
api/ai-insight/index.js  → Azure OpenAI proxy (~155 lines)

4.2 Data Binding

Every input field in the workbook carries an emid attribute that maps to an Encompass field ID. The SSF Guest API framework provides bidirectional binding — changes in the workbook push to the loan file, and changes from other Encompass screens reflect in the workbook immediately.

Zero Dependencies: SmartAdvisor™ uses no frameworks (React, Vue, etc.), no bundlers (Webpack, Vite), and no npm packages. This eliminates supply-chain attack surface and simplifies deployment to a static file copy.

5. AI Integration: Azure OpenAI

SmartAdvisor™ calls the /api/ai-insight endpoint, which is an Azure Function running within Azure Static Web Apps. This server-side proxy holds the API key and forwards requests to Azure OpenAI's chat completions API.

5.1 Prompt Engineering

The system prompt instructs the AI to return structured JSON with categorized insight objects. Each insight contains a title, detail, severity (risk/compliance/info), category, and array of relevant fieldIds.

// Request (client → Azure Function)
POST /api/ai-insight
{
  "prompt": "Analyze a $520,000 conventional purchase, 80% LTV, 762 credit, 35.86% DTI"
}

// Response (Azure Function → client)
{
  "insights": [
    {
      "title": "Loan Amount Within Conforming Limits",
      "detail": "$520,000 is within the 2026 conforming limit...",
      "severity": "info",
      "category": "compliance",
      "fieldIds": ["1003", "1005"]
    }
  ]
}

5.2 Privacy Architecture

Key Security Properties:
• API key stored in Azure SWA Environment Variables (server-side only)
• Client never sees, stores, or transmits the key
• Azure OpenAI runs in the customer's own Azure subscription
• Microsoft DPA guarantees data is not used for model training
• No PII (names, SSN, addresses) included in AI prompts — only financial ratios

6. Workflow Engine

The Workflow Engine (workflow-engine.js, 3,267 lines) is a client-side calculation and simulation core. It provides:

→ Milestone Tracking — 7 stages from Started → Completion with field-level readiness scoring
→ AUS Simulation — DU, LP, and Total Scorecard emulation with agency guideline checks
→ TRID Compliance — LE/CD timing calculations with waiting period enforcement
→ Conditions Engine — 23 condition types with auto-matching to eFolder documents
→ Pre-Audit — Mirrors Encompass loan audit for catch-before-submit validation

7. Competitive Comparison

SmartAdvisor™ is unique in combining AI intelligence with deep LOS integration — directly inside the loan form, not as a separate application.

Capability	SmartAdvisor™	Encompass Native	Third-Party POS
AI-powered insights	✓ Azure OpenAI	✗	~ Some
Real-time SSF binding	✓ Bidirectional	✓ Native	✗ Import/Export
Client-side AUS sim	✓ DU/LP/Total	✗ Server-side only	✗
Conditions tracking	✓ + eFolder match	~ Basic list	✗
Role-based workflows	✓ 6 roles	~ Personas	✗
TRID date engine	✓ Auto-calc	~ Manual	✗
Field navigation	✓ Click-to-field	✗ Manual search	✗
Dark mode	✓ Full	✗	~ Some
Zero dependencies	✓ Vanilla JS	✓ .NET native	✗ React/Vue
Deployment	✓ Azure SWA	✓ ICE hosted	~ Varies

8. Deployment & Infrastructure

SmartAdvisor™ is deployed as static files to Azure Static Web Apps with a single Azure Function for the AI proxy. Total infrastructure cost is minimal — SWA Free Tier covers up to 100GB bandwidth/month.

Deployment Pipeline:
git push → Azure DevOps → CI/CD Build → Azure SWA Deploy → CDN Edge (global)

Infrastructure Components

☁️

Azure Static Web Apps

Hosts all HTML, CSS, JS files. Managed functions for /api endpoints. Built-in CDN. HTTPS by default.

🤖

Azure OpenAI Service

gpt-4.1-mini deployment in East US. Customer tenant isolation. Standard S0 tier.

🔧

Azure DevOps

Git source control. CI/CD pipeline. Pull request workflow. Build validation.

🧪

Playwright Testing

Automated UI tests. Panel quality validation. AI integration tests. Cross-browser coverage.

9. Roadmap

Timeline	Feature	Impact
Q1 2026	Rate lock advisor — AI-powered rate lock timing recommendations based on market data	Reduce lock extension costs
Q1 2026	Gemini server-side proxy — Move Gemini key server-side for parity with Azure	Eliminate all client-side keys
Q2 2026	Multi-borrower intelligence — Separate AI analysis per borrower/co-borrower	Better risk stratification
Q2 2026	Automated condition clearing — AI verifies documents against condition requirements	Reduce processor workload 40%
Q3 2026	Predictive close dating — ML model trained on historical milestone data	Accurate close date forecasting
Q3 2026	Compliance rule engine — TRID/RESPA/ECOA rule automation with auto-alerts	Reduce compliance risk

SmartAdvisor™: AI-Powered Loan Intelligence
for Encompass Web

Contents

1. Abstract

2. The Problem: Fragmented Loan Intelligence

3. The Solution: SmartAdvisor™

4. Technical Architecture

4.1 Module Structure

4.2 Data Binding

5. AI Integration: Azure OpenAI

5.1 Prompt Engineering

5.2 Privacy Architecture

6. Workflow Engine

7. Competitive Comparison

8. Deployment & Infrastructure

Infrastructure Components

9. Roadmap

SmartAdvisor™: AI-Powered Loan Intelligencefor Encompass Web

Contents

1. Abstract

2. The Problem: Fragmented Loan Intelligence

3. The Solution: SmartAdvisor™

4. Technical Architecture

4.1 Module Structure

4.2 Data Binding

5. AI Integration: Azure OpenAI

5.1 Prompt Engineering

5.2 Privacy Architecture

6. Workflow Engine

7. Competitive Comparison

8. Deployment & Infrastructure

Infrastructure Components

9. Roadmap

SmartAdvisor™: AI-Powered Loan Intelligence
for Encompass Web