Chen
arrow_back All Work

Designing Horizon Bank Developer Central

Accelerating the developer lifecycle through modular architecture and self-service tooling.

Client anonymised under NDA.

ROLE Senior Product Designer
TIMELINE 4 Months
TEAM Team of 2

Problem

Engineering tools, documentation, and compliance processes were scattered across fragmented internal sites, creating a steep learning curve and constant support load on lead engineers.

Solution

A single developer portal organised around four tasks — Discover, Build, Learn, Manage — with interactive architecture diagrams, sandbox editors, and code generation built in.

Impact

Onboarding lead time dropped 65%, from weeks to days, and integration bugs fell 35%.

Project Overview

Horizon Bank's internal tools, docs, and compliance processes were scattered across disconnected sites — a constant source of friction for thousands of internal engineers. Developer Central consolidates all of it into one self-service platform.

Objectives

Methodology

I ran a five-phase, evidence-first process (below), validating every architectural decision against real developer workflows rather than assumptions.

psychology 1. Empathise
design_services 2. Define
lightbulb 3. Ideate
brush 4. Design
science 5. Test

What I Learned, and What I Had to Unlearn

Two things this project changed for me — one I picked up, one I had to put down.

What I picked up

How to coordinate independent front-end pieces inside a bank, where compliance has an opinion about nearly everything. And how to design for an audience that will take a plain, useful screen over a good-looking one every time.

What I put down

I was used to designing whole pages and controlling exactly how they looked. Here the page gets assembled from pieces owned by different teams, and it will never look precisely as I drew it. Getting comfortable with that took most of the four months.

1. Empathise: Understanding the Developer Context

Management had a theory about what was wrong. I wanted to see it, so I sat with 10 developers across seniority levels and watched them do ordinary things: find an API, set up an environment, work out what depended on what. Watching beats asking here — people forget the workarounds they've stopped noticing.

Survey Data Highlights (N=150)

Upgrade Readiness Awareness100%
Primary Communication for SupportOutlook / Teams
Technical Adoption Rate (Reference App)80%
Interest in Internal Community Forums71%

Key Pain Points Identified

Developer's Work-Life Ecosystem Details

Two numbers made the case: developers waited up to three weeks for environment configuration and documentation access before writing a first API call, and 20% of code failed in staging purely from local environment variance.

2. Define: Establishing Target Profiles and Journeys

Two people kept showing up in the research, and they wanted opposite things. I wrote them down so every later decision had something concrete to argue with.

Ramya

Intermediate Developer

Ramya needs step-by-step guidance to set up her environment and understand how MFEs work — clear "Getting Started" docs and error messaging matter most. Her goal: a faster "Time to First Hello World."

Ramesh

Senior Architect

Ramesh needs fast, unencumbered access to raw API specs, version control switching, and performance analytics. His goal: maintain system integrity while increasing team velocity.

The Developer Journey

I mapped the journey from "Project Assignment" to "First Successful Deployment." The legacy path spanned 14 touchpoints across 5 systems with manual approval gates; the redefined journey collapsed this into one portal with automated provisioning and contextual documentation.

Current-State User Journey: The Developer Experience

Macro-Phase 1: Set up & start building
Step 1
Discovering COSMOS
Actions

Rely on word-of-mouth or stumble upon the COSMOS Design system Home page.

Friction: Major gap in initial discoverability. Only found if already in use.
Step 2
Onboarding
Actions

Meet Product team, get access, training, read FAQs and architecture docs.

Friction: Raising access requests from multiple sources is highly time-consuming.
Step 3
Getting Started
Actions

Interact with Cosmos, API docs, set up Dev env, MFE Generator, Open hours.

Friction: Steep learning curve; reliant on "Open hours" support.
Macro-Phase 2: Building products
Step 4
Development

Navigate resources (JS/React guides), manage upgrades & security.

Friction: No ready-made codes, low discoverability of reference app.
Step 5
Testing

Conduct Unit Testing and raise defects to Library team.

Friction: Raising defects is unstructured, increasing resolution times.
Step 6
Production

Pass code through the team's automated quality checks (internally called the "FED Scorecard") before it's considered production-ready.

Step 7
Maintenance

Version upgrades, public release updates, community contribution.

warning Systemic Journey Blockers
  • cancel Fragmented Information: Repetitive content across various URLs.
  • cancel Lack of Sequencing: Difficult to understand correct steps to follow.
  • cancel Zero Self-Service: Dependent on "Open hours" support.
  • cancel Poor Cross-Discoverability: Siloed information.
  • cancel Missing MFE Support: Little to no MFE-specific material available.

3. Ideate: Structuring the Solution Architecture

Then the question became which of these to actually build. I ranked ideas on two things only: could engineering realistically ship it, and would it remove a step from someone's day.

The 'Big Ideas' Matrix

Automated Starter CodeA simple visual tool for generating starter components in one step, already wired to the company's security standards and style rules.
Integrated Sandbox EditorsEmbedding a live code editor directly in the documentation, so developers can test API calls and try out components without first setting up a project on their own machine.
System Architecture DiagramsVisualizing how systems connect with clear, interactive diagrams, helping developers see what depends on what before making changes.
Switching Between VersionsAn always-visible switch that lets senior engineers move between API versions and documentation states without losing their place in the portal.

Information Architecture & Flows

I cut the site structure right back. Instead of organising by internal team or org chart, I organised around what developers were actually trying to do. The primary navigation was reduced to four verbs: Discover (APIs & Components), Build (Tools & Starter Code), Learn (Documentation & Patterns), and Manage (Infrastructure & Access). I tested the structure with senior engineering leads before anything got built.

4. Design: From Wireframes to High-Fidelity Execution

This audience notices when you patronise them. The interface had to carry a lot of information without feeling dumbed down — engineers read dense screens all day and don't thank you for hiding things.

Wireframing Core Utilities

Low-fidelity wireframes established a modular grid for dense data tables, code snippet blocks, and multi-pane navigation. The core goal: get a new user from landing page to starter-code download in under three clicks, validated in iterative reviews with engineering.

Visual Design & System Integration

The visual system stayed monochromatic, with brand color reserved for interactive states and status indicators, and a single type scale — Inter — across both UI and long-form documentation for a consistent reading experience. I documented interaction behavior for modals, nested dropdowns, and code blocks, with spacing and structural annotations precise enough for pixel-perfect handoff.

Before and after of the Developer Central homepage. The first version listed what existed; the revision leads with search, an explicit guided tour for newcomers, a what's-new strip, direct routes to raise a defect or request, and counters for available components, use cases and MFEs

The first build answered "what exists here". Testing showed new developers were asking something else first — "what do I do now" — so the revision leads with search and a guided path instead of a product catalogue. Screens redrawn with client branding removed.

Three linked screens: a numbered four-step getting-started path, the component index grouped by buttons, communicators, data tools and form components, and a single component page carrying live expandable code and its accessibility rules

Setup was rebuilt as a numbered, finite path, and every component page carries copyable live code plus its accessibility rules — so the documentation answers the question at the point the developer has it.

Documentation & Playground Solution Details

Documentation had the same problem as the rest of the platform: scattered across PDF specs and Confluence with no live playground, so developers deployed to staging just to inspect a payload response. An integrated live sandbox let them validate API contracts in real time instead.

5. Test & Results: Project Impact

Post-launch validation combined quantitative analytics with qualitative feedback loops built into the new portal.

Metric AreaLegacy System (Before)Design (After)
Search & DiscoveryFragmented across 3 platformsUnified Central Hub
Onboarding TimeManual, mentor-dependent (Weeks)Self-service guided paths (Days)
Technical UtilityStatic, disconnected documentationInteractive code gen & integrated editors
Onboarding lead time↓ 65%
Integration bugs↓ 35%

FED Scorecard metrics improved alongside these. The larger shift was architectural and cultural: from a page-building mindset to a node-based, interoperable one — and treating internal developers with the same rigour as external users, which turned a utility people avoided into a platform teams chose.

The hardest constraint was serving both ends of the Ramya-to-Ramesh spectrum in one interface — simplifying enough for a first-time user without stripping away the raw control power users like Ramesh needed, which meant almost every screen went through a round of "is this too much, or not enough."

Phase 2 — Extending Developer Central with AI

Status: Piloted. The figures in this section are projected, based on pilot data — not yet measured at full production scale.

Phase 1 solved centralisation. But research surfaced a deeper problem: developers weren't blocked because information didn't exist — they were blocked because finding it, and converting it into working code, took real effort. The next step wasn't better documentation navigation. It was AI embedded in workflows developers already used, rather than another standalone tool.

Research Insights

Our generative research revealed two recurring pain points across beginner and experienced developers.

Insight 01

"I need ready-made code to start faster."

New developers spent considerable time configuring their environment, understanding COSMOS dependencies, and manually writing repetitive starter code before getting to the actual business logic.

Insight 02

"I know the documentation exists. I just can't find it."

Resources were distributed across COSMOS, API documentation, GitHub repositories, migration guides, and internal learning platforms. Developers spent more time searching than building.

Design Opportunity

The two highest-impact opportunities to evolve Developer Central from a documentation portal into an intelligent development assistant were:

Solution 01

AI-Assisted MFE Code Generation

Project setup was the single most painful moment for new developers — exactly where Insight 01 hit hardest.

Our Design Approach

Instead of creating another AI application, we enhanced the existing MFE Generator already available inside Developer Central. Developers drag UI components from the COSMOS design system onto a visual canvas, and the AI generates ready-to-use code for them — properties, imports, and dependencies already wired up — ready to check against the team's existing quality standards (the FED Scorecard).

Design Principle

Rather than replacing developers, AI removes repetitive setup work while keeping developers in control of implementation. Developers spend time solving business problems—not writing repetitive starter code.

80% Projected reduction in project setup time
65% Projected reduction in Level-1 support requests
40% Projected MFE adoption increase within one quarter
Solution 02

Smarter Search & Discoverability

Research consistently showed that developers already knew the documentation existed. The real challenge was knowing where to look. Traditional keyword search required developers to understand document names, repositories, and platform structure before they could search.

Our Design Approach

Rather than designing new search UI, we replaced the underlying search engine with one that understands the meaning behind a question, not just its keywords. The existing global search bar became an AI-powered knowledge assistant.

search "How do I change a button to Standard Skin in React?"

Instead of returning dozens of matching files, Developer Central now provides a direct answer with the relevant code snippet, links to the exact COSMOS component, and related migration guides.

Retrieves knowledge from:
COSMOS Component Library API Registry Migration Documentation Release Notes
65% Projected reduction in search time
Self-Service Resolve implementation questions independently
Unified Knowledge One source for all framework knowledge

Workflow Comparison

Before

Search Docs
Open Multiple Platforms
Read Documentation
Ask Support
Write Starter Code
Start Development

≈ Several hours to days

After

Open Developer Central
Ask AI / Generate MFE
Receive Working Code
Validate with FED Scorecard
Start Development

≈ Minutes

User Feedback

"Earlier I had to wait for someone to help me set up COSMOS. Now I simply drag the components I need, and the generator creates everything for me. I can start building immediately."

— Beginner Developer

"Instead of searching across multiple platforms, I ask a question and immediately receive the exact implementation with links to the correct documentation. It finally feels like one developer platform."

— Senior Developer

"Routine onboarding questions have reduced significantly. Instead of answering repetitive setup issues, we can now focus on improving the platform itself."

— Platform Support Team

Reflection

This phase moved Developer Central from somewhere you look things up to something that helps you do the work.

The clearest guardrail was scope: the generator stayed limited to the repetitive starter code every project needs, with architecture and business-logic decisions always left to the developer — widening that scope carelessly would have been the fastest way to erode trust in the tool.

Next Case Study
Fintech DeFi / DLT Equity Design

Bridging the Financing Gap for DOSBs

Fair and equitable capital access through Distributed Ledger Technology. Dismantling systemic barriers for diverse-owned small businesses.

Senior Product Designer
arrow_forward