PREP. The Pierce Real Estate Platform.

Context

Real estate teams run on spreadsheets. Generic CRMs don't fit how transactions actually work. The market has had this problem for a decade. Existing tools either cost too much or solve the wrong problem.

I started seeing it through my wife's network. She's in real estate. Her coworkers were managing dozens of transactions in a spreadsheet that one person owned. When that person took a Friday off, the whole shop slowed down.

That's not a process. That's a single point of failure with a branding problem.

Approach

Built PREP from the ground up for the actual workflow. Not "CRM with a real estate skin." Shaped around how a transaction actually moves: pipeline, offer, escrow, close.

Stack:

• · Next.js 15 App Router · React 19 · Tailwind v4
• · Supabase + Postgres + Prisma
• · Vercel + AWS for infrastructure
• · Stripe for billing
• · Vercel AI SDK + Anthropic Claude for AI features

Architectural calls that mattered:

• Multi-tenant from day one. Brokerage holds teams, teams hold agents, agents work with TCs. The data model never had to be retrofit.
• Field lifecycle pattern. Every data field traces from input through API through DB through audit log. Data is silently lost when any layer is missing.
• Audit-first design. Real estate is regulated. Deletes leave traces. Every change is attributable.
• Mobile is first-class. Agents live in their car. Desktop is the secondary surface.

What's actually built:

• Transaction tracking. Pipeline lanes per stage. Per-deal checklists. Role-based handoff between agent, TC, and broker.
• CRM. Contacts, communication log, lead source attribution.
• Automated workflows. Documents generated on milestone. Notifications when deals stall. Inspections scheduled automatically.
• AI-assisted follow-up. Claude drafts personalized lead-follow-up emails. Agent reviews and sends. Conversion, not chasing.
• Team collaboration. TC hands a deal back to the agent without a four-message Slack chain.
• Subscription gating. Tiered access by brokerage size.

Outcome

Live with paying customers. The beta cohort came through my wife's network. Actual agents and TCs using it for real transactions. The feedback loop is tight. I ship a fix the same day someone hits a bug.

The system has handled deals end-to-end: pipeline lane to offer accepted, escrow opened, inspection scheduled, docs signed, close.

Reflection

Building real estate software is harder than it looks. The domain has 50 years of conventions and a thousand edge cases. What happens when a buyer's agent and listing agent are at the same brokerage? What about dual agency? What about TCs outside the brokerage? Every feature shipped with a "but..." question I hadn't anticipated.

The thing I'd do over: spend more time with TCs early. They're the ones in the spreadsheet all day. Agents can hand off. TCs can't.

What I'd keep: the boring decisions. Postgres over a graph database. Server-rendered React over a SPA. Stripe over a custom billing system. The shiny choices would have cost six months I didn't have.