Martita

Martita is a WhatsApp-native AI assistant for Mexican SMEs — the receptionist a small business can’t afford to hire. I designed, built, and run it solo: one codebase, multi-tenant, fully isolated per business, live with paying customers at martita.ai.

What it does

A customer messages a business on WhatsApp in messy, multi-intent Spanish — “hola, tienen cita mañana? y cuánto cuesta el corte?” — and Martita handles it the way a human receptionist would: answers questions about the business, books and reschedules appointments, and remembers returning customers across sessions.

How it’s built

• Agentic core: Claude reasons over 21 tools (scheduling, catalog, customer memory, business config, escalation) instead of a rigid intent-classification pipeline. Multi-intent messages are handled naturally because the model plans across tools.
• Multi-tenant by design: one deployment serves every business, with hard per-tenant isolation of data, configuration, and conversation history.
• Cross-session memory: returning customers are recognized and context carries over — the difference between a bot and a receptionist.
• Self-serve onboarding: a business signs up, pays via Stripe, and configures itself without me in the loop.

~7,500 lines of code, built and operated by one person. Martita is also my proving ground: production agent patterns get battle-tested here before I bring them to enterprise work at GM.