#Build01 · Restaurant tech · AI analytics

AI analytics platform

One AI layer replaces disconnected POS and QR dashboards. Owners see what happened, why, and what to do next.

/month saved

faster decisions

fewer people to manage data

Problem

Owners, CFOs, and managers each read their own dashboard. Rista POS and DotPe QR data lived apart. Margin leaks and stock outs went unnoticed until the revenue was already gone.

Solution

Horizon merges POS and QR data into one AI insight layer. It ranks issues by severity, attaches a specific fix to each one, and lets an owner send that fix straight to WhatsApp in one click. A conversational panel lets them ask the data anything directly.

Process

Built the pipeline first: nightly re-aggregation into 7/30/60/90 day windows, two Python scripts per window compressing each slice into a sales and customer briefing, then one Claude call per window turning those briefings into ranked insights. The dashboard reads only from precomputed views, so it never waits on AI.

Technical details

How it's built.

What runs where, what it talks to, and what was handed over.

Stack

PythonClaude Sonnet API (claude-sonnet-4-6)PostgreSQL materialized views

Integrations

Rista POSDotPe QRWhatsApp Business API for one-click insight actions

Delivered

Overview, Revenue Pulse, Menu IQ, Customer Engine, and Channels modules
Conversational query panel over merged POS and QR data
Nightly, severity-tagged insights with a recovery estimate on each
One-click insight-to-WhatsApp action loop
Rolling 7/30/60/90 day insight windows, regenerated daily without blocking the dashboard

Architecture

How the pieces connect.

Raw data is aggregated daily, then sliced into 7/30/60/90 day windows. Parallel Python scripts compress each window into briefing JSON. Both briefings feed one Claude call per window, which returns ranked insights.