Built two AI products and deployed them in our real estate company in five weeks

ACHIEVEMENTS:

- THE MAIA SUITE — we built and deployed a family of AI "digital employees" for our own real estate company, designed from day one to be sold to other developers: Maia Venta (sales) and Maia Obra (construction site).

- MAIA VENTA — much more than a chatbot: it's our ENTIRE commercial strategy, redefined and re-planned EVERY WEEK from all sales information sources — the AI agent's WhatsApp conversations, in-person visits, funnel numbers, corresponde and ad performance. Every Monday, next week's commercial actions are defined based on what actually happened last week: which messages converted, which channel produced, which leads to reactivate, at what price per typology. The execution layer runs on GoHighLevel, fully integrated: project websites (viveginkgo.com and tuproyectovive.com), Facebook/Meta campaigns, and a WhatsApp agent (built on ChatGPT within GHL) handling buyer conversations end-to-end — TRY IT LIVE: +34 624 271 894 (https://wa.me/34624271894). A strategy that learns weekly, executed by a workforce that never sleeps. THE ENGINE BEHIND IT — two custom Claude skills power that weekly redefinition: the first codifies Daniel Priestley's "Find, Understand and Win Your Best Customers" methodology (from this very program) — Short Form → Long Form → Lead Form → LAPS pipeline (Leads→Appointments→Presentations→Sales), affluent-niche ICP. The second, "Vive Ginkgo Commercial Strategy", runs the weekly analysis against real data. Salespeople transcribe every in-person visit with Granola, and Claude coaches them after every single one, from their own words. Both skills live inside "ADN Investleading", versioned and governed in Airtable.

- MAIA OBRA — much more than an incident tracker: it's the ENTIRE construction site, digitized. Anyone taps the exact spot on any of the 46 real floor plans (served live from Google Drive) and the incident exists: auto-numbered, photographed, assigned to its trade and contractor, with a deadline. Generic defects spanning many homes, multi-trade incidents, "post-sale risk" flagging for future liability claims — everything our site actually deals with. Each profile sees what needs THEIR action ("4 incidents awaiting your validation — oldest: 6 days"), and every step is signed and dated: builder resolves with photo evidence → site director validates → architect validates → auto-close. Built by me — a CEO who cannot code — in 5 weeks, zero developers, live in production today (maiaobra.com secured). - THE CYCLE BEHIND IT — the app is eating the weekly site meeting: agendas ("Orden del Día") now generate themselves from open items — incidents flagged for the visit, points carried over meeting after meeting (with a counter so NOTHING gets forgotten), overdue items. Next: the site director dictates the visit to Claude, verdicts are tapped on the phone (✅ verified / ❌ rejected / ⏸️ carried over), and the weekly minutes write themselves on our own template — 32 historical meeting minutes already migrated as the archive. Traceability from construction to post-sale: when a buyer claims a defect two years from now, the full history — photos, dates, who validated what — is one click away.

- ARCHITECTED TO BE SOLD — Airtable is the backbone of the whole system: database AND configuration layer for both products (incidents, approval chains, catalogs, teams, commercial tracking). Everything that varies between clients is configuration, never code — onboarding a new developer means filling tables, not a development project.

- THE INVISIBLE ASSET — "NDA Investleading", our corporate brain: 49+ structured knowledge documents (investment criteria, commercial playbooks, sustainability standards, build methodology) written in Markdown, versioned in Google Drive, indexed in Airtable, with a self-improvement protocol: after meaningful work sessions, learnings are proposed, validated and versioned into the brain. Both commercial skills live inside it, governed like every other piece of company knowledge. This is what makes the suite replicable: a client doesn't just buy software — they buy a documented, continuously-learning way of running sales and site management that we operate ourselves every day. The suite's own build method is one of its documents (META_02: "How we build internal apps with AI").

- MARKET VALIDATED FROM INSIDE — we ARE the customer; first external pilot candidates identified in our own network of developers, builders and architecture studios.

TOOLS: - Shared backbone: Airtable (database + ALL configuration for both products) · Claude Cowork (strategy, specs, brand, infrastructure, knowledge capture) · "ADN Investleading" knowledge base (Drive + Airtable) - Maia Venta: GoHighLevel + ChatGPT (WhatsApp sales agent) · Granola (visit transcription) · Meta/Facebook & Google Ads · two custom Claude skills (Priestley methodology + weekly commercial analysis) - Maia Obra: Claude Code (100% of the code) · Next.js on Vercel (auto-deploy from GitHub) · Google Drive with service account (plans, photos, documents) · GitHub (private repo)

METHODOLOGY: - Maia Venta — doctrine + weekly rhythm: Priestley's system (Short Form → Long Form → Lead Form → LAPS) as fixed doctrine; then a weekly cycle: all sales data in (agent conversations, Granola visit transcripts, funnel KPIs, ad performance) → Claude analysis → next week's commercial actions defined Monday → salespeople coached after every visit. Strategy is never static: it's re-planned every single week from reality. - Maia Obra — spec-first building: Dream→Design→Develop spec → validate the riskiest piece with a one-screen prototype → build in "buckets" that only close when real users (our site director, architect and builder) have tested them → numbered feedback log becomes the next consolidated build order → release checklist before every deploy. - Shared golden rules: everything that varies is configuration in Airtable, never code · AI proposes, humans decide (validations, liability, investor communications) · every learning is captured and versioned into the company brain. - Roadmap to revenue: (1) finish both pilots on our own project, (2) self-generating weekly meeting minutes, (3) multi-tenant + login, (4) first external pilots with developers from our own network, selling the suite.

LEARNINGS: The best B2B products come from operators who automate their own pain. AI collapses the cost of validating a venture — we got production pilots BEFORE spending a euro on developers, inverting the classic startup risk curve. In sales: a strategy re-planned weekly from real conversations beats a perfect annual plan — the market tells you what works every seven days, if you have a system that listens. But the deepest learning is about learning itself: our entire Claude setup is DESIGNED to learn — every week, learnings from sales, site and strategy are proposed, validated and versioned into our skills and our corporate brain (each skill carries its own self-improvement protocol). The company doesn't just use AI; it compounds knowledge weekly. Two rules govern everything: AI proposes, humans decide (validations, liability, investor communications) — and everything that varies is configuration, never code. Oh, and naming matters: "Maia Obra" plays on "mano de obra" (workforce) — the marketing writes itself.

RELEVANT URLs: Live pilot (Maia Obra): https://app-incidencias-obra-omega.vercel.app · Commercial sites with Maia Venta integrated: https://viveginkgo.com · https://tuproyectovive.com · Try the agent on WhatsApp: https://wa.me/34624271894 ·