AI
AI Automation
We run the automation steps that need judgement with AI.
The situation
Classic automation works on fixed rules. When incoming documents or requests arrive in a different shape each time, writing rules gets hard.
Our approach
We hand the variable part to the model. Classification and routing run on AI, and the steps after them continue on rules.
What we do
- Classifying incoming requests
- Extracting data from documents
- Automatic routing and prioritisation
- Confidence thresholds and human handover
- Tracking result accuracy
Technologies we use
- Backend
- Node.js · Express · PHP · Payload CMSNode.js speaks the same language as the frontend in most projects; we use PHP where only shared hosting is available.
- Database
- PostgreSQL · Supabase · SQLite · MySQL · Prisma · DrizzlePostgreSQL for multi-user, relational work; SQLite for small, single-site setups.
- Servers & hosting
- Linux (Ubuntu) · Nginx · PM2 / systemd · Let's EncryptA dedicated Linux server is enough for most projects and keeps costs predictable; server access is handed to you.
- AI
- Claude API · Gemini APIWe pick the model by the sensitivity and cost of the task, and state in writing where your data goes.
How we work
- 01
Discover
A map of the current state and a prioritised list of needs.
- 02
Define
A scope document, success criteria and a schedule.
- 03
Design
A clickable interface prototype and the design system.
- 04
Develop
Working builds delivered at regular intervals.
- 05
Test
Test scenarios, a closed bug list and an acceptance test signed off by you.
- 06
Deploy
The live system, technical documentation and a hand-over session.
- 07
Grow
A usage report and a prioritised improvement plan.
Common questions
What happens when the model is not confident?
The work goes to a person. We set the confidence threshold together and anything below it goes for approval.
Other services in this area
Let's talk about this
A short call to get clear on what you need. We will tell you plainly which approach fits.
Start your project