Multi-tenant AI Assistant for Mortgage Lending Guidelines

Loan officers lose hours searching long, frequently updated agency and investor guideline documents, and every lender layers its own overlays on top. Each new lender used to mean another separate front end.

Developer on the platform. Built admin and user dashboards and the AI chat experience, and worked on its tenancy model: the tenant is resolved from the domain, branding and feature flags live in configuration, and every record is scoped to its tenant so lenders' data never mixes.

What the product does. Details are kept generic to respect the client's confidentiality.

• Database-driven bot workflow engine (overlay, parallel and summarizer steps) that queries several guideline sources at once and streams one combined answer.
• Self-serve broker workspaces: admins create assistants from uploaded PDFs and manage users, trials, add-ons and token usage.
• A structured review of an investor guide across 48 attributes that flags conflicts, takes corrections and exports a workbook.
• Stripe subscriptions with free-tier limits, an embeddable chat widget, voice input and an AI newsletter pipeline.

One platform where single-client front ends became tenants: 250+ API handlers, 96 pages, 40 data models and 480+ automated test cases, with 17 scheduled jobs running billing, retention and guideline-update checks.

• Next.js 16
• React 19
• TypeScript
• tRPC
• Prisma
• PostgreSQL
• NextAuth
• Stripe
• OpenAI
• Sanity
• AWS
• Vitest