The market

$4T industry. $30B in commissions. Zero modern AI.

The category is dominated by Encompass (ICE), Calyx, and Empower (Fiserv) — all 20-year-old loan-origination systems built before broadband, let alone AI. None has a credible AI product. None can ship one fast: they're locked into legacy architectures and post-merger integration cycles.

The window for an AI-native mortgage platform is open right now — and it won't stay open long.

The problem

Loan officers spend 60% of their day on lookups.
Only 40% with clients.

A loan officer's job is supposed to be advising a family through the biggest financial decision of their lives. In practice, the day looks like this:

• ×Hunting through 100+ lender rulebooks — each one hundreds of pages, updated quarterly
• ×Phone tag with 5–10 account execs per deal, just to get answers that should be searchable
• ×Drafting emails, comparison letters, underwriter memos by hand — every loan, from scratch
• ×Switching between 8+ disconnected tools — none of them talk to each other

The result: families lose deals they should have won. LOs burn out. Brokerages can't scale past the founder's personal capacity.

The product

Cal — the AI underwriter that ships today.

Cal is the brain at the center of the platform. A loan officer types a question in plain English. Cal answers in 30 seconds — with the right lender, the matching guideline, the AE contact, and a citation to the source document and effective date. Live in production. Used in real deals.

• ✓1,424 rulebook documents indexed and searchable
• ✓166 wholesale lenders with current contact info
• ✓250 down-payment-help programs across all 50 states
• ✓Cites the source for every answer — no guessing, no hallucination
• ✓Refreshes its own library every month, automatically

A real example

Here's what it actually looks like.

A loan officer types this question. Cal answers in seconds.

Without Cal, this question takes 45 minutes of phone tag with three different lenders. With Cal, 30 seconds.

The data moat

The library is the hardest part to replicate.

A general-purpose AI can't answer mortgage questions accurately because the source data lives in unindexed PDFs, gated broker portals, and AE phone calls. We've spent three months building the ingestion pipeline, the categorizer, and the freshness rules that make this library trustworthy — plus an automated monthly refresh that keeps it current.

This isn't web-scrapable. Every other AI mortgage tool starts at zero.

Cal is just the brain

Around it, a full stack of products.

Cal is the most visible piece, but it's one node in a platform we've shipped over the last few months. Every piece is live in production today.

• ·White-label LO portals — every loan officer gets a fully branded version of the entire stack: their photo, their colors, their phone, their voice. Same brain underneath.
• ·Realtor portal — co-branded for partner real estate teams. Pre-approval letters, deal tracking, down-payment-help eligibility check, open-house tools, automatic notifications.
• ·Founder dashboard at command.homeloanexpress.ai — live cost monitoring, chat analytics, library health, every team member's activity in one view.
• ·DPA Assistant — consumer-facing tool that walks any borrower through 250 down-payment-help programs in 50 states.
• ·Reverse Mortgage Engine — Clay's HECM tool. One-click flyers, scenario math, native iOS app on the way.
• ·Document Drop tool — drag any paystub, tax return, 1003, or bank statement onto the page. Cal auto-classifies, extracts the data, and pushes it into the loan file.
• ·Email + letter generator — audience-tailored tone (borrower, realtor, underwriter, account exec) and a Cal-built underwriter memo per file.
• ·Rate Forecast Studio — six-factor lock-signal scoring, trained on FRED + RSS + MBS data.
• ·Public changelog + blog system — the build is shipped in the open. Every release is logged.

Each of these used to be a separate product some company would charge thousands of dollars a month for. We built them as one connected system — because the loan, the borrower, the realtor, and the underwriter are all part of the same job.

Why this matters

Tech does the looking-up.
Humans do the listening.

A mortgage is the biggest debt most families will ever take on. The person guiding them through it shouldn't be buried in 100-page rulebooks, lender phone trees, and document-processing busywork. They should be present — listening, understanding the family's real goals, naming the trade-offs honestly, making the borrower feel safe in a process that is built to make people feel small.

Cal handles the lookups, the comparisons, the underwriter memos, the AE coordination, and the document classification. The loan officer handles the part that requires being a person — care, judgment, and compassion. That trade is the whole point.

Where this goes next

Cal becomes an agent.

Today Cal answers questions. Soon it will act — across the loan officer's entire workflow, autonomously, while the LO is on the phone with the family.

Step one shipped today, May 8. We open-sourced cal-mcp-server — the first mortgage MCP server with a real broker-accessible lender library behind it. MIT-licensed, public on GitHub. Cal is now reachable from Claude Desktop, Cursor, Continue, and every other MCP-compatible AI client. MCP (Model Context Protocol) is Anthropic's open standard for AI agents — and we just made Cal the canonical mortgage layer for it.

Step two: Cal as MCP client. Through MCP, Cal will reach into the CRM, the loan origination system, the pricing engine, email, calendar, e-sign, and more. From a single instruction, Cal will:

• →Pull the live rate sheet for the borrower's scenario
• →Run the qualifying math against the top three lenders in parallel
• →Draft the comparison letter and the AE intro email
• →Schedule the follow-up call on the LO's calendar
• →Push the structured loan summary into the CRM
• →Flag any compliance gap before the file is touched

All from one prompt. All while the LO is being a human for the family in front of them.

Traction

Real loan officers. Real questions.
Real numbers.

The four pilot loan officers cover a deliberate spread of segments — not random users:

• ·Hector Ramirez — Silicon Valley + Elk Grove · purchase volume + Spanish-speaking market
• ·Diana — Bay Area · first-time homebuyer specialist
• ·Mike — Bay Area · new-construction + builder relationships
• ·Clay — Reverse mortgages · senior demographic, growing fast

Every one of them was an existing loan officer using legacy tools who switched to the platform within their first 90 days of seeing it. Acquisition channel: founder-direct. Conversion: 100%. Retention so far: 100%.

Why we win

Three moats compounding.

• 1Data moat — 1,424 lender guidelines indexed, categorized, freshness-tagged. Behind broker-portal logins, not on the public web. Months of work to replicate, and getting deeper monthly via automated refresh.
• 2Distribution moat — the founder is an active mortgage broker (Chris Black, NMLS 275073, 17 years). LOs are recruited inside the industry through trust and warm intros, not paid acquisition. Acquisition cost so far: ~$0.
• 3Workflow moat — Cal isn't a chatbot bolted onto someone else's stack. It's integrated across the LO's entire day: chat, package generator, email + letter, doc drop, dashboard, branded portals. Once an LO is in, switching cost is high.

The legacy POS players (Encompass, Calyx, Empower) can't catch up. They're locked into 20-year-old architectures, mid-acquisition, and distracted. The pure-AI startups have no domain knowledge, no LO distribution, and no data — they start at zero on day one.

Business model

Three revenue streams. 80%+ gross margin.

• 1Per-seat SaaS — $100–200/mo per loan officer. Standard B2B SaaS pricing model.
• 2Brokerage enterprise tier — $5K–20K/mo per brokerage for white-label deployment, custom branding, admin controls. Five-figure monthly contracts.
• 3HomeLoanExpress origination revenue — every loan funded on the platform generates broker comp directly to the parent company. Internal LOs (4 today, 100+ planned) compound this.

Unit economics

The math already works.

SOM math: 1,000 paying LOs × $150/mo = $1.8M ARR. 10,000 LOs × $200/mo = $24M ARR. We are six to eighteen months from the first milestone, and the second is a credible 36-month line of sight given the founder-direct acquisition channel.

Why now

The window is open — and it won't stay open.

• →AI inflection — frontier models (Claude 4, GPT-5) are finally accurate enough to handle structuring questions with citations. 18 months ago this product wasn't possible.
• →MCP just shipped — and we're first in mortgage — Anthropic's open agent protocol (2024) is the rail the next wave of vertical AI agents runs on. Today we open-sourced cal-mcp-server, the first mortgage MCP server with a real lender library. The mortgage-industry MCP layer is ours to maintain — and incumbents (Encompass, Calyx, Empower) have nothing.
• →Incumbent vacuum — Encompass (ICE acquisition), Empower (Fiserv), Calyx are all mid-integration cycle. None will ship credible AI in the next 24 months.
• →Market downturn = LO consolidation — the bottom 30% of LOs are exiting. The remaining 70% will adopt anything that compounds their productivity. Perfect demand-side moment.

The round

$500K in. $1.5M committed.
Closing the rest now.

Use of funds: 40% engineering + AI build · 30% LO acquisition + onboarding · 20% sales + content · 10% infra + ops. Target end-of-runway: 100 paying LOs, >$1M ARR run-rate.