← Back 02 · Product Experience Case Study

AI Engine Integration

Taking previously simulated AI features and making them real — I researched the integration architecture, built the backend logic in Claude Code, and defined the integration criteria with engineering, compressing a typically longer integration into roughly two weeks.

Role

Design & engineering lead (design + IT collaboration)

Context

Enterprise B2B SaaS · connecting a code-based prototype to a live AI engine

Duration

~2–2.5 weeks (in progress)

Stack

Claude Code · code-based prototype · AI engine integration · backend logic

Confidentiality note: product and client names are withheld. This project is in progress at time of writing; described as an active initiative.

01 · Situation

The company's marketing and sales code-based prototype demonstrated AI features convincingly — but those features were simulated. The data and AI behaviour were reproduced in the prototype layer, not powered by a real engine. To move from a compelling demo to functional product, the prototype needed to connect to a live AI engine so the AI features would work for real.

Before

Simulated AI features

Data and AI behaviour reproduced in the prototype layer

→

After

Live AI engine

The AI features work for real

This is normally a longer, engineering-owned integration. Here it ran as a lean design-and-IT initiative — no Business Analysis layer, design working directly with engineering.

02 · Task

  • Research the technical approach for connecting the prototype to a real AI engine — integration architecture, technical details, and the shape of the link.
  • Build the backend logic required for the integration.
  • Define the integration criteria jointly with IT.
  • Complete the code steps required to make the link functional.
  • Do it on a compressed timeline, in parallel with other running projects.

03 · Action

1 · Deep technical research

Researched possible integration solutions in technical detail — how the link between the prototype and the AI engine should be architected, what the data flow needed to look like, and what the trade-offs were.

2 · Backend logic in Claude Code

Built the backend logic for the integration using Claude Code — moving from research into working implementation rather than stopping at a specification.

Integration architecture — prototype to live engine
FRONT Code-based prototype UI + AI features MINE · CLAUDE CODE Integration layer Backend + data flow EXTERNAL Live AI engine Real inference request ↓ ↑ response call ↓ ↑ result INTEGRATION CRITERIA · SET WITH IT What "done" means, agreed up front PROD KEYS · IT In progress · step 5/5 No Business Analysis layer — design talks to IT directly

3 · Criteria-setting with IT

Collaborated with IT to define the integration criteria — the shared rules and conditions the link had to satisfy — so both sides were aligned on what "done" meant before finishing the technical work.

4 · Final integration steps

Completed the remaining code steps required for the link. IT is now handling the production keys and the remaining technical layer.

Production keys & remaining technical layer

With IT · in progress

04 · Result

Status register4 of 5 steps complete
Backend logicBuilt Integration criteriaDefined Production keys & final technical workWith IT Projected close~2–2.5 weeks

(In progress — projected close ~2 to 2.5 weeks, extended slightly by parallel projects.)

  • The initiative is moving positively and on track to convert previously simulated AI features into functional ones.
  • The backend logic and integration criteria are built and defined; production keys and final technical work sit with IT.
  • The integration is being delivered in a shorter window than this kind of work typically requires.

05 · What this project shows

  • A designer operating as a design engineer — researching integration architecture and building backend logic, not just designing the interface.
  • Direct design-to-engineering collaboration without an intermediary layer.
  • Use of AI-assisted development (Claude Code) to move from research to working backend implementation quickly.
  • Ownership of the boundary between design and production — turning a demo-grade prototype into a functional AI product.
← All work Next: Tirvela Boutique →