Reducing content discovery friction

I led the re-design of content discovery within Brix, a set of enhancements to enable users to discover and reuse machine learning code assets, snippets, visualisations, notebooks, and pipelines, more quickly, reducing time-to-value for AI innovations.

Surface

Web App

My role

Design

Timeline

Q3 2023 – Q1 2025

Skills

Product Design, User Research & Testing, Design Systems

The Challenge

As Brix grew to 1,500 MAU and 700+ code assets, finding resources became increasingly difficult.

66% of users abandoned filter searches
Users started filtering but gave up mid-flow, overwhelmed by too many options and an unclear information architecture.

Filters behaved inconsistently across contexts
Filter options and behaviour differed between Posts and Collections, forcing users to relearn the pattern for each content type.

Multiple navigation layers consumed viewport space
The sidebar, header navigation, and topic pills competed for attention, leaving minimal room to preview assets. Users couldn't assess relevance without clicking, forcing blind exploration.

Audit and review of the old UI

Slide 1 of 2

Research & Solution

A review of successful filtering and search systems revealed four core principles essential to solving Brix's content discovery problem.

A review of successful filtering and search systems revealed four core principles essential to solving Brix’s content discovery problem.

Transparency

Users see the filtering logic behind every choice

During discovery, three filtering approaches were tested and the 'filters popover' emerged on top, balancing impact with feasibility.

During discovery, three filtering approaches were tested and the ‘filters popover’ emerged on top, balancing impact with feasibility.

The filters popover emerged as the right choice for this phase, not because it was the easiest solution, but because it addressed the immediate concerns. While the agentic search offered a more ideal user experience, where users could simply describe what they're working on and the system would surface relevant assets, testing revealed that users valued control and transparency, they wanted to see how assets were structured rather than rely on a black box. Rather than leap to an unproven, resource-intensive solution, I prioritised fixing what was broken and delivering immediate value. The agentic search became a roadmap priority for the next phase.

The filters popover emerged as the right choice for this phase, not because it was the easiest solution, but because it addressed the immediate concerns. While the agentic search offered a more ideal user experience, where users could simply describe what they’re working on and the system would surface relevant assets, testing revealed that users valued control and transparency, they wanted to see how assets were structured rather than rely on a black box. Rather than leap to an unproven, resource-intensive solution, I prioritised fixing what was broken and delivering immediate value. The agentic search became a roadmap priority for the next phase.

Implementing a grid structure and card redesign made filtered results instantly scannable and actionable

Filters solved the first part of the problem: helping users narrow down options. But research revealed a second friction point: once filtered, users still struggled to evaluate results quickly. The cognitive load shifted from "too many options" to "unclear what I'm looking at."

Filters solved the first part of the problem: helping users narrow down options. But research revealed a second friction point: once filtered, users still struggled to evaluate results quickly. The cognitive load shifted from “too many options” to “unclear what I’m looking at.”

Redesigned asset cards for faster differentiation

Content-type badges, short descriptions, key tags, and engagement signals so users can judge relevance at a glance, cutting decision time and reducing unnecessary clicks. Consistent visuals across cards also reinforced that each result is the same kind of asset.

Restructured the results grid with a consistent system

A grid system with consistent spacing and alignment allowed users to scan predictably. It also surfaced applied filters clearly, so users always understood their current context.

Impact

In the 12 months since shipping the feature, the new filtering system has transformed how users discover reusable code assets

47%

filter searches completed

Up from 34%. Designed to stop people giving up on filters and falling back to browsing.

52%

found an asset on first try

Up from 41%. Designed to put relevant assets in front of users sooner.

Learnings

Fix what’s broken, before chasing what’s shiny

The AI chat was the more exciting idea, but research showed people wanted control over how they searched. Choosing the filters popover meant shipping something that helped straight away, and it gave the team a solid foundation to build agentic search on later.

Reflections

What i'd do differently:

Separate the impact of each change

Discovery and the article page shipped close together, so staging releases would make each one’s effect clearer.

Prototype the chatbot cheaply

A quick test would have shown how much control people really wanted, with less guesswork.

What i'd do next:

Agentic search, with filters alongside

Let people describe what they’re working on, while keeping filters visible for those who want control.

Learn from zero-result searches

They show what users need that Brix doesn’t have yet, which is a prompt for new contributions.

RECENT PROJECTS

VIEW ALL PROJECTS

Shipped

Simplifying information architecture

How a page redesign increased code asset contributions by 48%

Shipped

Simplifying information architecture

How a page redesign increased code asset contributions by 48%

Shipped

Simplifying information architecture

How a page redesign increased code asset contributions by 48%