NotChatbot
AI CRM for ecommerce. I own the product's metrics and built Views: smart dashboards you create by talking to an AI.
- Role
- Full Stack Developer · Metrics owner
- Period
- Oct 2025 — now
- Stack
- Next.js · Convex · TypeScript · Vercel · LLMs
- Link
- notchatbot.com ↗
Demo with sample data, recreated from the product.
01 — The problem
Brands have thousands of conversations a day, and understanding them can't depend on a fixed report.
NotChatbot handles support and sales over WhatsApp, Instagram, Mercado Libre, email and web chat. The data was there, but every new question (which channel sells most? how much does the AI solve on its own? where do conversations get stuck?) was a request to the team. Fixed metrics answer yesterday's questions, not today's.
02 — What I did
Views: ask the AI and it builds the dashboard.
- Views: users ask in plain language for what they want to see and the AI builds a dashboard with KPIs, charts and tables
- Dashboards that are saved, refined by chatting (“add last month”) and refresh themselves with new data
- The product's metrics layer: the definitions, calculations and queries behind Views and reporting
- Full-stack features with AI and ecommerce integrations, in Next.js and Convex
03 — Key decisions
AI builds the view, the system computes
The model turns the question into which metrics to show and how. Numbers always come from queries, never from the model.
One definition per metric
Each metric is defined once and reused in Views and reports: the same number means the same thing everywhere.
Dashboards you talk to
Instead of a chart editor, you refine the dashboard by asking. Business people don't need to be data people.
04 — Outcome
I own NotChatbot's metrics.
Views went from idea to product feature, and everything measured in NotChatbot goes through the metrics layer I maintain.