AI pipeline tool

Internal tooling for building purpose-driven AI systems

Role: Product & UX Lead
Duration: 2 months, part-time
Product: Internal AI configuration platform
Focus: Workflow clarity, cognitive load reduction, iteration speed

The configuration screen: the task list on the left, the selected task's settings in the middle and a live test chat on the right.

Overview

This project redesigned an internal platform for building and managing AI chat systems for very specific uses. The prompting system underneath was powerful and precise. The tool for configuring it had become slow, complex and hard to reason about.

At Strata, I evaluated the existing experience and reworked the workflows and the interface so people could reach the system's capabilities faster and more easily, without losing control or flexibility.

Problem

The technical foundations were strong, but the tool held people back:

Managing complex prompt chains and dependencies took a lot of mental effort
Common configuration tasks were inefficient
It was hard to see how prompts flowed through the system
Testing and refining behaviour was slow

The users knew the domain. The tool made everyday work harder than it had to be.

Key decision

I fixed the workflow before touching the UI.

With two months, part-time, a full rebuild was off the table. I restructured how configuration works first: clear stages, prompt flows and dependencies made visible, and fewer steps for the common tasks. The visual refinement came after.

Approach

The principles:

Restructure workflows before refining the UI
Make system behaviour and dependencies explicit
Cut the steps for common tasks
Reveal complexity only when it is needed

The aim was an interface that matches how people actually think about building and refining AI behaviour.

The organizations list, showing each one's runs this month and its state.

Solution

The redesign gave the tool a clearer structure and faster workflows.

Streamlined configuration

Clear stages for configuration and iteration
Visible structure around prompt flows and dependencies
Shorter paths for common actions
Better hierarchy, feedback and awareness of state

I used high-fidelity designs and interactive prototypes to test the new approach with stakeholders.

A new configuration at its starting point, with one task and the test panel ready.

Outcome

Bots are faster to create and revise
Complex logic is easier to manage
People can see how a change will affect the system's behaviour
Users are more confident working with live configurations

Discuss this case study with you?

I'd be happy to walk through this project in more detail and discuss how this approach might apply to your needs.

Set up a call (opens in a new tab) Contact me