Selected work

Applied AI

Restaurant Menu Assistant

Applied AI that runs on the restaurant's own hardware.

The Restaurant Menu Assistant is a self-hosted AI system for restaurants. It pairs a kiosk- and tablet-friendly chat interface that helps diners find dishes by preference or dietary need with an admin panel for managing items, categories, multi-currency prices, discounts, and images. It uses retrieval-augmented generation over the restaurant's own menu and runs entirely on the local network after setup.

  • Retrieval-augmented generation
  • Self-hosted AI deployment
  • On-premise data privacy
  • Cost-conscious model choice

The context

Restaurants want AI help for diners without shipping menu data to the cloud or paying a per-query fee for every question asked. This work addresses how to deliver useful, private AI on hardware a venue already owns.

What we delivered

  • A customer chat interface for kiosks and tablets that answers menu questions and finds dishes by preference or dietary need
  • An admin panel to manage items, categories, multi-currency prices, discounts, and images, with automatic menu reindexing after edits
  • A retrieval-augmented answering pipeline using Ollama with qwen2.5:3b for chat and bge-m3 for embeddings, backed by Qdrant semantic search
  • A self-hosted deployment via Podman / Docker Compose with PostgreSQL, MinIO image storage, an Nginx proxy, and token-authenticated admin
  • Operational discipline: explicit memory budgeting for 8GB hosts, health checks, automated backup and restore, per-install secrets, and LAN-only defaults

Why it matters

  • Keeps data private and on-premise, with no per-query cloud cost
  • Shows retrieval-augmented generation applied judiciously and cost-consciously, with documented model tradeoffs
  • Demonstrates the operational rigour to run AI reliably on modest, customer-owned hardware
  • The same approach can bring private, offline AI to any business wary of cloud dependence

Capabilities shown

  • Retrieval-augmented generation
  • Self-hosted AI deployment
  • On-premise data privacy
  • Cost-conscious model choice

Built with

  • Python
  • TypeScript
  • Ollama
  • qwen2.5:3b
  • bge-m3
  • PostgreSQL
  • Qdrant
  • MinIO

Where is the value hiding in your business?

Let's have a direct, no-pressure conversation about where AI and technology could actually move your numbers.

Book a conversation