Skip to content
Bhushan

Command menu

Jump to a page or system

All systems

Restaurant intelligence platform

Live website

BhuSutra

Smarter Restaurants. Greater Growth.

A restaurant and cafe platform that covers point of sale, kitchen display, tables, inventory, purchasing, staff, expenses, billing, reports and a profit & loss statement — with an analyst that answers questions about the restaurant's own numbers.

Product website is published at getbhusutra.com. The application itself runs on the owner's own computer.

Problem

Restaurants generate data in many separate places: sales, stock, purchases, wages, costs. Reports built from them are usually late, and answering a simple question means exporting and combining spreadsheets.

Approach

Keep everything in one local database, build the reporting a finance person actually needs (profit & loss, break-even, cost health), and add simple, explainable models on top instead of a black box. The aim is data → understanding → prediction → action, not just data → reports.

Features

Reports that filter each other
Business overview with a what-if growth planner, profit & loss, sales, purchases, inventory with a reorder planner, and menu timeline. Ranges compare with the previous period and export to Excel.
Ask page
Questions are matched to known intents by vector similarity using a small local embedding model — no LLM and no API call. Quick facts are computed from order history with DuckDB.
Explainable ML layer
Per-item margin, rolling-average demand, bought-together co-occurrence, variance and anomaly detection, procurement and staffing recommendations.
Anomaly checks with tests
A test plants realistic problems (supplier price rise, electricity spike, heavy discounting, waste) and checks the Watchtower finds each one while a clean copy stays quiet.
Two-step sign-in
Google account first, then a passkey or an authenticator-app code, with one-time recovery codes, hashed session tokens, encrypted secrets and lock-out after repeated wrong codes.
Local-first data
One SQLite file on the machine. Nothing is uploaded; only Google's sign-in check touches the internet.

Next system

Project Reincarnation

Read case study