Product design and platform
CheckMyRef
A platform for storing, managing and reusing employment references, with a brand built around clarity and trust.
Read the case studyVisit product (checkmyref.com, opens in a new tab)
AI product engineering
AI that belongs in the product, not beside it.
Phigz builds software where model behaviour, product logic and the interface are designed together. The result is a product a business can operate, not a chat window bolted onto an unfinished app.
The problem
Many teams have tried a model in a prototype and then stalled. The demo answers questions. It does not fit the workflow, the data, or the way a customer actually finishes a task.
The other failure is the opposite: a conventional product with an AI label. A text box is added late, the rest of the system is unchanged, and nobody owns what the model is allowed to do.
AI product engineering is the work of deciding where a model helps, where rules should stay deterministic, and how a person reviews the result before it reaches a customer or a record system.
What Phigz delivers
Product definition
The job to be done, the user, the data you already have, and the decision the software should support. Scope stays narrow enough to ship.
Model boundary
What the model may read, what it may suggest, and what it must never write on its own. Prompts are treated as product behaviour, not as a one-off experiment.
Application around the model
Interface, API, persistence and permissions so the AI feature works inside a real product rather than a notebook.
Review and release
Test cases for ordinary paths and for awkward inputs, then a deployment path the team can repeat.
Where it fits
A feature inside an existing product
Search, drafting, classification or recommendations added where users already work, without a rewrite of the whole platform.
A new AI-native tool
A focused product whose main value is intelligent assistance, still backed by ordinary accounts, data and admin controls.
An internal product for a team
Software staff use every day: briefing, triage, summarising a queue, or preparing a decision for a person to confirm.
Delivery
01
Frame the decision
We write down the user, the input, the output and the cost of a wrong answer before choosing a model.
02
Shape the product
Interface and system design come before prompt tuning, so the model has a clear job.
03
Build the path
Application code, data access and the model call are implemented as one feature, with logs a person can read.
04
Prove it, then ship
We review behaviour on realistic cases, fix what fails, and release behind a path you can operate.
Technical capability
Application stack
Web products are typically built with TypeScript, React and Next.js or a comparable modern stack, with APIs in Node.js, Laravel or PHP where that fits the existing system.
Model integration
Calls stay on the server. The product defines the schema of the response, timeouts, and what happens when the model is unavailable.
Deployment
We ship to cloud environments teams already use, including AWS, Google Cloud and Azure, without presenting those as partnerships.
Related product work
Published case studies are product builds. Where a page is about AI, these projects show how Phigz ships a complete product. They are not labelled as AI systems.
Booking product
Test Dates
A booking product for learner drivers looking for earlier DVSA test dates, with availability monitoring and a mobile-first interface.
Read the case studyVisit product (testdates.co.uk, opens in a new tab)
Training platform
Bluetech Consulting
A training and career-support platform, designed around course discovery and a visual identity in professional blue.
Read the case studyVisit product (bluetechconsulting.com, opens in a new tab)
Questions
Do you only wrap a public chat model?
No. A public model may be part of the design, but the engagement covers the product around it: permissions, data, interface, failure behaviour and release.
Can this start as a small feature?
Yes. A single well-bounded feature is often the right first release. We would rather ship one path that people use than a broad AI programme.
Will you guarantee model accuracy?
No. Models are probabilistic. We design review, constraints and tests so the business can see how the feature behaves and decide what is acceptable.
Next step
Talk through this engagement.
Tell us the product, the current state, and the outcome you need. We will reply with a straight view of whether this is the right shape of work.