Moe Loubani
Back to projects
Buildventory
Next.jsMCPFastifyPostgreSQL

Buildventory

Operations Platform for New-Home Builders

Jul 2026

Drop your real price sheets and AI onboarding seeds your account with your actual inventory. Web, API, and MCP run the same audited commands, so an agent can do anything a user can.

An operations platform for new-home builders covering inventory, deals, holds, commissions, and options and upgrades, with AI-driven onboarding that seeds your account from your real price sheets.

Visit Buildventory

Overview

Buildventory is an operations platform for new-home builders: inventory (projects, units, floorplans), deals with deposit schedules and worksheets, holds and reservations with expiry, commissions, an options and upgrades configurator with a rules engine, and an availability board that draws the building as a stacking elevation with keyboard-navigable unit chips. A unit can never be double-sold. The onboarding is the standout part: AI onboarding is the acquisition flow. Drop a real price sheet or unit tracker, CSV, Excel, or PDF, and a worker pipeline parses it with per-field confidence scores, asks a few gap questions, and seeds your account with your actual data.

Technical Challenges

  • Onboarding from arbitrary price sheets means parsing wildly inconsistent spreadsheets and PDFs into a consistent data model, and knowing which parsed fields to trust versus flag for a human to confirm.
  • Letting an AI agent seed a live account with real data demands the same audit trail and guardrails a human action would get, not a shortcut around them.
  • A stacking-elevation availability board with keyboard-navigable unit chips has to stay accurate in real time as holds expire and units move between available, held, and sold.
  • Every command in the system, whether triggered by a person or an agent, needs to run through the same validation, policy, and audit path, without maintaining two parallel code paths for “human” and “AI” actions.

Solution

  • A Turborepo monorepo: Next.js for the web app, Fastify for the API, a dedicated MCP server, and a BullMQ worker for the parsing pipeline, all on Postgres 17 and Redis.
  • A single command registry holds all business logic, each command defines its Zod input schema, policy, and handler, and is exposed identically through the REST API and an OAuth-protected MCP server, so an AI agent can do exactly what a user can do, nothing more.
  • Every command execution is recorded as an ActivityEvent with full provenance, AI-issued commands are tagged actorType ‘ai’, with read-only versus full-access controls to scope what an agent is allowed to touch.
  • The onboarding worker pipeline parses uploaded CSV, Excel, or PDF price sheets and unit trackers field by field, attaching a confidence score to each, then asks the builder a short set of gap questions before seeding the account.
  • Better Auth handles authentication with 2FA/TOTP, an org switcher, and a command palette, deposit schedules, holds with expiry, and commissions run on the same command registry as everything else.

Key Features

  • AI onboarding as the acquisition flow: drop a price sheet or unit tracker (CSV, Excel, PDF) and get a seeded account, not a blank one
  • Worker pipeline parses uploads with per-field confidence, then asks gap questions before seeding
  • Single command registry powers both the REST API and an OAuth-protected MCP server, so an agent can do anything a user can
  • Every action recorded as an ActivityEvent with full provenance, read-only vs full-access controls
  • Availability board draws the building as a stacking elevation with keyboard-navigable unit chips
  • A unit can never be double-sold
  • Options and upgrades configurator with a rules engine
  • Deposit schedules, worksheets, holds/reservations with expiry, commissions
  • Free to start, plans from $99/month