AI workflow and prompt design
A defined job for the AI, the inputs it uses and the output it must produce.
Software & AI development
AI features and automation
I build AI features and automation that add practical AI to a product or workflow, such as knowledge retrieval, document processing, assisted conversations or controlled agent workflows. I build each feature around a measurable job, with evaluation, human review and cost controls designed in from the start.
Discuss this service ↗Is this right for you?
Who this is for, and who it isn't.
A good fit if
- A repetitive task involves reading, summarising, classifying or drafting
- Customers or staff struggle to find answers in your own documents and data
- You want AI in your product but need it to be reliable and affordable
Probably not the right fit if
- The task needs guaranteed, exact results with no room for review: conventional software is usually better
- You are unsure the idea is viable: start with a Project viability review
What you get
A practical result, not just a conversation.
Knowledge, tool or CRM integration
Connecting the AI to your documents, data and systems.
Evaluation and quality checks
Tests that show whether the feature is working, and catch regressions.
Usage, cost and safety controls
Budgets, limits and review points so the feature stays affordable and safe.
How it works
Three stages, agreed up front.
Define the job
You and I agree the task, the success measure and where human review is needed.
Build and evaluate
I build the feature and test it against real examples before release.
Release and monitor
I release it with cost and quality monitoring in place.
Evidence
Relevant work you can check.
Published projects and guides that show this kind of work in practice.
FewzenAI WhatsApp agents case study
A platform for goal-oriented WhatsApp agents using OpenAI, with campaign management, CRM integration and interaction analytics. Teams could automate customer conversations while retaining control over agent behaviour and campaigns.
Read the case study ↗Automotive finance CRM
Internal CRM integrating ten existing APIs, with AI-assisted conversations. Supported more than 2,000 production deployments in 18 months and trained over 60 employees.
Read the case study ↗Managing AI costs (free guide)
The cost-control approach used in these engagements, published in full: unit economics, cost ledgers, model choice and budget alerts.
Read the guide ↗Common questions
Before you get in touch.
How much does it cost to add AI to our product?
Builds are scoped in stages. I agree a first stage with a fixed scope and price with you, then plan the next stage with you based on what the first stage has taught us. That keeps cost predictable without pretending every requirement is known on day one. Running costs are estimated before the build, and budgets and alerts are part of the delivery.
How do you stop the AI getting things wrong?
You cannot eliminate errors entirely, so the design limits their impact: a narrow job, grounding in your own data, evaluation against real examples, and human review where mistakes would matter.
Is our data sent to AI providers?
Usually some data is, so I agree what can be shared with you, choose providers and settings that meet your requirements, and keep secrets and sensitive data out of prompts where possible.
Work with me
Is this the right place to start?
Tell me what you are trying to achieve and you and I can work out the most useful next step. Start with a 30-minute call with me.