AI Chatbot Implementation for Restaurant Group

Client Profile

A major UK-based restaurant group operating 60+ locations, struggling with information fragmentation across disparate manuals (SOPs, HR, Compliance, Recipes).

The Challenge

Operational inconsistency and risk were high due to "document chaos." Critical answers regarding compliance (e.g., Natasha's Law), HR (Right to Work), and Safety were buried in dense, static PDFs. Frontline staff could not access the "single source of truth" during service, meaning critical checks were often ignored.

The Solution

Knowledge Graph-based AI Chatbot (Proof of Concept).

Structured Knowledge: Transformed unstructured manuals into a Neo4j Graph Database, linking Roles (e.g., Head Chef) directly to Responsibilities and Risks.

Semantic Search: Implemented a RAG (Retrieval-Augmented Generation) pipeline using a custom Python build and Gemini, allowing staff to ask natural language questions and receive precise citations.

Mobile-First Strategy: Recognized that a traditional web application creates friction for manual processes—such as documenting equipment faults or taking inventory for provisioning orders. By leveraging the "phone in your pocket" or tablet devices, the solution democratized access, ensuring the system is usable by all employees, not just management restricted to back-office terminals.

Key Outcomes

From "Never" to "Instant": Transformed information retrieval time from "never" (ignored files) to instantly provided on personal phones.

Comprehensive Operational Scope:

All Staff: Streamlined access for schedule requests, vacation planning, training modules, and in-house employment opportunities.

Management: Centralized control over the vendor network, accounting and projections, scheduling, hiring pipelines, POS integration, and training oversight.

Compliance Certainty: Provided auditable, consistent answers across all 60+ locations.

Scalability: Proven architecture capable of ingesting new manuals automatically without code changes.

Project Status

This engagement concluded as a successful Proof of Concept. While the prototype demonstrated all required technical capabilities and operational efficiencies, the client ultimately did not proceed with the production build. I was not privy to the internal strategic decision-making process regarding the project's discontinuation.

Technologies Used

Neo4j (Graph Database)

Custom Python Architecture

Google Gemini API (LLM)

RAG Architecture