Back to Home
Project 01

MaxiCLI

Multi-tenant SaaS · WhatsApp × AI

A full platform that turns WhatsApp into an AI-powered commerce and customer-service channel — admin dashboard, automated orders, analytics, and an internal AI assistant for operators.

Role
Solo architect & lead engineer
Status
Built to production-grade
Period
2024 — 2025
Domain
Conversational commerce
FastAPINext.js 15PostgreSQLpgvectorRedisWebSocketRAGMulti-tenant
The Problem

Small commerce teams running on WhatsApp lose orders to chaos: messages scattered across phones, no audit trail, no analytics, no automation. Off-the-shelf chatbots are rigid; building from scratch needs a long runway.

Approach

Architect a strictly-isolated multi-tenant SaaS, an agent runtime that proposes actions but never executes silently, and a 35+ table schema engineered to evolve without painful migrations.

Architecture

Tenancy & data

PostgreSQL with row-level scoping per tenant, pgvector for retrieval, Redis for queues and ephemeral state. Alembic migrations tied to spec docs.

Agent layer

Dual-model setup — a fast model for routing and extraction, a stronger model for reasoning. Tool-calling with idempotency keys and structured pending actions.

Realtime

WebSocket fan-out to the admin dashboard, with optimistic updates and re-sync on reconnect.

Quality

220+ test files covering domain rules, schema constraints, and agent contracts before they reach production.

Outcomes
  • Reached production-grade across the full stack.
  • Multi-hour operations collapsed to a single confirmation.
  • High engineering velocity from clean, explicit layering (domain / service / repo / API).
Stack
FastAPINext.js 15PostgreSQLpgvectorRedisRAGWebSocketsAlembicDocker
Screenshots
Admin dashboard — orders, analytics, and per-tenant control.
Operator AI assistant — proposes structured actions, never executes silently.