Quality management with AI

QAissist

Quality work in childcare went from binders, spreadsheets and lost reports to one system that notices when action is needed and starts an improvement cycle with a proposal.

role
Lead developer
stack
Laravel Livewire + Volt Flux UI Laravel Reverb Laravel Horizon RAG / AI integration
Qaissist
context

What was going on

QAissist is a quality management system for childcare, built to commission to replace scattered binders, spreadsheets and lost reports. It brings reports, PDCA improvement cycles, KPIs, document management, standards and regulations together in one system and does not wait: it notices when action is needed and starts an improvement cycle with a proposal from QAISSIST AI. A built-in assistant answers staff questions from the organisation's own documents and the law, with sources.

challenges

What made it hard

Streaming AI answers in real time through queued workers and WebSockets, through a tightly secured production environment

Watertight separation between organisations, combined with visibility per location and role down to menu level

Keeping automatic improvement cycles across all modules manageable with a declarative workflow engine of its own

A reliable connection to the AI layer: signed messages, webhooks and a knowledge base that syncs exactly once

Following laws and regulations automatically, with an estimate of the impact per organisation

approach

How I approached it

Built server-first on the TALL stack: Livewire and Flux UI for the interface, Horizon queues for all the heavy work and Reverb for streaming AI answers in real time. A declarative workflow engine with reusable steps runs the automation between modules, from approved report to completed improvement cycle. Since go-live I host and maintain the platform.

result

Live in use at organisations in childcare

Extended to a second sector: construction and installation

Around 1,700 automated tests, with explicit tests on the separation between organisations

A platform for many organisations with licence tiers and an AI layer under its own name

intake

Need something like this?

One route: analysis, building and introducing, and maintenance if you want it. It starts with a thirty-minute intake.