largelanguage.mortgage

An independent, buyer-first AI mortgage advisor. No lender affiliation. No conflict of interest. The mortgage AI that works for you, not the bank.

Why This Exists

Every consumer-facing mortgage AI tool (Better.com's Betsy, lender chatbots) is owned by a lender — built to sell you their product. There is no neutral, intelligent, conversational AI that explains your specific situation honestly, including when renting might be the better choice. This is that tool.

Features (MVP)

• AI Mortgage Advisor — Conversational AI that answers any mortgage question in plain English. Streaming, session-based, no login required. Supports Claude, GPT-4o, local models, and BYOK.
• Local LLM Support — Connect to your own Ollama or vLLM deployments with custom model names and endpoints.
• Scenario Calculator — Side-by-side mortgage scenario comparison, amortization schedules, points break-even, rent vs buy. Runs entirely client-side, zero latency.
• Live Rate Tracker — Weekly FRED/Freddie Mac rates with AI-generated plain-English commentary on what moved and why.
• Waitlist — Email capture for the owned audience (newsletter, future features).

Stack

Layer	Technology
Frontend	Vue 3 + TypeScript + Vite → static build → Cloudflare Pages
Backend	NestJS + TypeScript + TypeORM + Postgres + Redis + BullMQ
AI routing	LiteLLM proxy (Claude, OpenAI, Ollama, BYOK)
Infra (prod)	Nomad + Consul + Vault + Traefik
Monorepo	pnpm workspaces

Project Structure

largelanguage.mortgage/
├── apps/
│   ├── web/                  # Vue 3 + Vite frontend (Cloudflare Pages)
│   └── api/                  # NestJS backend
├── services/
│   └── litellm/              # LiteLLM proxy config
├── infra/
│   ├── nomad/                # Nomad job specs
│   ├── consul/               # Service mesh config
│   ├── vault/                # Secret policies
│   └── traefik/              # Routing rules
├── docker/                   # Dockerfiles for prod images
├── docs/                     # Phase implementation guides
├── docker-compose.yml        # Local dev services
├── .env.example              # Required environment variables
├── CLAUDE.md                 # AI agent instructions
└── package.json              # pnpm workspace root

Local Dev Quickstart

Prerequisites

• Node.js 20+
• pnpm 9+
• Docker + Docker Compose

1. Clone and install

git clone <repo>
cd largelanguage.mortgage
pnpm install

2. Configure environment

cp .env.example .env.local
# Edit .env.local — add ANTHROPIC_KEY (minimum), FRED_API_KEY

3. Start backing services

docker-compose up -d
# Starts: Postgres (5432), Redis (6379), LiteLLM proxy (4000)

4. Run apps

# Two terminals:
pnpm --filter api dev     # NestJS on :3000
pnpm --filter web dev     # Vite on :5173

5. Verify

curl http://localhost:3000/health
# → { "status": "ok", "postgres": true, "redis": true }

Environment Variables

See .env.example for the full list. Minimum for local dev:

Variable	Description
DATABASE_URL	Postgres connection string
REDIS_URL	Redis connection string
LITELLM_URL	LiteLLM proxy URL (default: http://localhost:4000)
ANTHROPIC_KEY	Anthropic API key (for LiteLLM)
FRED_API_KEY	Federal Reserve FRED API key (free at fred.stlouisfed.org)

In prod, all secrets come from Vault — see infra/vault/.

Implementation Phases

Phase	Doc	Sub-phases
1 — Foundation	phase-1-foundation.md	Monorepo, docker-compose, NestJS + Vue bootstrap
2 — Core Features	phase-2-core-features.md	2a Waitlist · 2b Calculator · 2c LLM · 2d Chat
3 — Rate Tracker	phase-3-rate-tracker.md	3a FRED Client · 3b BullMQ Jobs · 3c Commentary · 3d Rates View
4 — Deploy	phase-4-deploy.md	Human engineer guide · DevOps Agent Guide

Using Local LLMs (Ollama/vLLM)

The application now supports connecting to your own Ollama or vLLM deployments.

How to Use:

1. Select "Local (Ollama)" from the model dropdown
2. Enter your model name (e.g., "llama3", "mistral", "gemma")
3. Enter your Ollama/vLLM endpoint URL (default: http://localhost:11434)

Configuration Details:

The system uses LiteLLM's built-in support for Ollama models. Your custom configuration will be passed through to LiteLLM, which will route requests to your local deployment.

Local Development Setup:

If you're running Ollama locally, make sure:

• Ollama is running on your machine
• The endpoint is accessible from the Docker container (use host.docker.internal for local development)
• Your models are pulled and available on the Ollama instance

Data Sources

• Mortgage rates: FRED API — MORTGAGE30US, MORTGAGE15US, MORTGAGE5US series (Freddie Mac PMMS, published weekly on Thursdays). Free, no scraping.

License

Private / proprietary.