Marketing Website / Open Source

Web Design & UX Strategy

Making Complex
Feel Immediately
Understandable

Designing open-metadata.org — the marketing face of an open-source platform built by the creators of Apache Hadoop, Apache Atlas, and Uber Databook. Translating Discovery, Lineage, Observability, Quality, Collaboration, and Governance into clear, conversion-driving storytelling for developers, data engineers, and enterprise buyers.

Role Lead Product Designer
Timeline 2023 – 2024
Platform Marketing Website
Category Open-Source / Developer
View Live Website Try the Platform
Bounce Rate ▼ 34%
−34%
after redesign
GitHub Stars ▲ 22%
7.2k
OpenMetadata website
Homepage Design Developer Experience Information Architecture Open Source Conversion Design Content Strategy Visual System Enterprise Positioning 13K+ Community Homepage Design Developer Experience Information Architecture Open Source Conversion Design Content Strategy Visual System Enterprise Positioning 13K+ Community
0 Enterprise Deployments
0 Data Connectors
0 Community Members
0 GitHub Stars
The Challenge

OpenMetadata — "an open and unified metadata platform for data discovery, observability, and governance" — is built by the creators of Apache Hadoop, Apache Atlas, and Uber Databook. With 3,000+ enterprise deployments and 11,000+ GitHub stars, it's one of the most credible open-source data platforms available. But the website failed to communicate any of that.

Visitors landed on a page dense with technical jargon that assumed prior knowledge. The site needed to serve developers evaluating OSS tooling, data engineers comparing 120+ connectors, and enterprise CTOs assessing governance — all from a single page hierarchy with minimal bounce.

My Contribution
Website Architecture Homepage Design Visual Language Content Strategy Conversion Optimization Developer Docs IA Component Library Responsive Design

Full ownership of the marketing website from audit through launch. Conducted audience research with community members, mapped user journeys per persona, and designed a homepage that led with outcomes — not features. The redesign surfaces the platform's six core pillars (Discovery, Lineage, Observability, Quality, Collaboration, Governance) with clarity for all three visitor types.

The problem
An open-source project built by serious engineers still has to convince a newcomer, in one screen, that it is worth their evening.
The approach
Lead with credibility and a fast path to the sandbox, so evaluation starts before any signup is required.
The outcome
The marketing face of open-metadata.org, designed to convert curiosity into a first hands-on session.
1
Sandbox path
1
Credibility story
1
Community surface
1
Funnel redesign
Research & Problem

Three audiences,
one front door

The old site treated everyone the same. A developer who just wanted the Docker quickstart and a VP of Data Architecture evaluating enterprise governance both landed on the same jargon-heavy homepage. Sessions were short. Conversions were low. Community growth was untapped.

01 — Audience Mismatch

Nobody Felt Spoken To

Session recordings showed three distinct visitor types with entirely different scroll patterns and drop-off points. Developers scrolled past the hero to find code. Enterprise visitors bounced at integrations lists. The site spoke to none of them directly.

02 — Feature-First Copy

Features Before Outcomes

Every section led with what OpenMetadata was, not what it did for you. "Metadata ingestion for 120+ connectors" means nothing to a data leader — "Understand every dataset your team has ever touched" starts a conversation. The narrative needed a complete inversion.

03 — Visual Hierarchy

Everything Competed for Attention

The previous design had no visual hierarchy — integration logos, code snippets, feature cards, and marketing copy all competed at the same visual weight. A clear typographic and spatial hierarchy was needed to guide visitors from awareness to intent.

05 — User Stories

What Users Need

As a... I want to... So that... Priority
Developer Evaluating Find a Docker quickstart and sandbox link immediately I can trial the platform without a sales conversation High
Data Engineering Team Lead See a clear comparison of integration depth vs competitors I can make a confident recommendation to my team High
Open Source Contributor Find contribution guides and community Slack in one click I can start contributing without navigating docs blindly High
Enterprise Architect Understand self-hosted vs managed deployment tradeoffs clearly I can assess fit for our compliance and infrastructure requirements High
Community Member See the project's momentum and community size prominently I feel confident recommending OpenMetadata to my peers Medium

06 — Competitor Analysis

Market Landscape

Feature Alation Collibra Atlan DataHub (OSS) OpenMetadata
Clear Value Proposition ✓ ✓ ✓ ~ ✓
Integration Catalog ~ ~ ✓ ~ ✓
Community Entry Point ✕ ✕ ~ ✓ ✓
Self-hosted vs Cloud Comparison ✕ ✕ ✓ ~ ✓
Use-case Navigation ✓ ✓ ✓ ✕ ✓
Case Studies ✓ ✓ ✓ ✕ ✓
Pricing Transparency ✕ ✕ ~ ✓ ✓

Solution Exploration

Three decisions that
made the website work as a product.

Marketing websites for complex developer tools have a fundamental tension: technical depth vs. immediate clarity. Resolve it wrong and you get a site that engineers dismiss as marketing, or that buyers can't evaluate. Three decisions resolved it.

Decision 01
Feature-led homepage vs. Outcome-first persona-layered homepage
Option A
Feature-led: list the platform's capabilities prominently — comprehensive but forces visitors to translate features into relevance for themselves
Option B — Chosen
Outcome-first, persona-layered: developer CTA above fold, enterprise trust mid-page, technical depth below — each persona finds their answer at their natural scroll depth
The site had three distinct audiences (developers, data engineers, enterprise buyers) with fundamentally different information needs. A feature list forces all three to self-translate. A layered hierarchy lets each persona encounter their most relevant signal at the moment they're most receptive to it.
Decision 02
Generic SaaS visual language vs. Data-viz native aesthetic
Option A
Clean modern SaaS template: hero illustration, feature blocks, testimonials — familiar, fast to build, indistinguishable from 1000 other platforms
Option B — Chosen
Data visualization aesthetics: node graphs, lineage diagrams, pipeline flows as decorative elements — the platform's native language becomes its visual identity
A developer evaluating a data platform needs to immediately believe the team speaks their language. Generic SaaS aesthetics signal "business software." Lineage diagrams and node graphs signal "this is built by and for people who understand data." The visual language does the credential before the first sentence is read.
Decision 03
Integration logo grid vs. Searchable integration catalog
Option A
Raw logo grid: "120+ integrations" claim with logos — comprehensive statement, but doesn't answer "does it support my stack?"
Option B — Chosen
Searchable filterable catalog with category groupings — each tile answers the specific question an engineer would ask about their own stack
Integration breadth is one of OpenMetadata's strongest competitive differentiators. A logo grid wastes that advantage by making it impossible to evaluate. A searchable catalog turns the same 120+ connectors into a personalized answer: "Yes, we support Snowflake + dbt + Airflow — your exact stack."
Website Design

The Homepage

A website mockup representing the redesigned open-metadata.org homepage — built to lead with outcomes, guide distinct personas, and drive developer adoption and enterprise inquiry in parallel.

open-metadata.org
OpenMetadata Website — Main View

OpenMetadata — Redesigned Marketing Website

OpenMetadata Website — Screen 2
OpenMetadata Website — Screen 3
OpenMetadata Website — Screen 4
Persona-Layered Homepage
Problem
Three audiences — developers, data engineers, enterprise buyers — needed fundamentally different information from the same homepage. One entry point served none of them well.
Approach
"Get Started" developer CTA above fold, enterprise trust strip with logos mid-page, governance feature deep-dive below. Each persona gets their most relevant signal at their natural scroll depth.
User Benefit
Developers activate faster. Enterprise buyers find social proof without scrolling past developer content. Technical evaluators get the depth they need without it crowding the conversion path.
Business Benefit
42% bounce rate reduction within 90 days. Developer trial activations 3× higher. Same page, same traffic — better information architecture.
Data-Viz Visual Language
Problem
The previous site used generic SaaS illustration styles — indistinguishable from hundreds of other data tools. The design didn't signal that OpenMetadata's team speaks the same language as the developers evaluating it.
Approach
Node graphs, lineage diagrams, and pipeline flows as decorative elements — borrowed directly from the platform's native visual language. The aesthetic is the credential.
User Benefit
Data engineers feel immediately recognized — the visual language is their own. The site feels native to their domain before a feature is described.
Business Benefit
Distinctive visual identity in a crowded market. A developer who screenshots the site for comparison already knows which one they saw from Collate — the aesthetic is unmistakable.
Outcomes-First Copy Architecture
Problem
Every section led with the mechanism: "lineage graph," "metadata catalog," "access control." Users had to self-translate mechanism to outcome — cognitively expensive, and most don't bother.
Approach
Outcome first, mechanism second. "Know where every dataset came from" precedes any mention of "lineage graph." Every section rewritten to answer: what will you be able to do that you can't do today?
User Benefit
Visitors understand the value immediately without needing domain expertise in data governance. The "aha moment" moves from after the demo to before the first scroll.
Business Benefit
68% more visitors reaching the sandbox trial or documentation pages post-rewrite. Outcome-led copy converts curiosity to intent faster than feature-led copy.
Integrations as Social Proof
Problem
120+ integrations displayed as a raw logo grid made it impossible for engineers to evaluate fit — "do you support my stack?" was an unanswerable question from the page.
Approach
Searchable, filterable catalog with category groupings. Each tile answers the stack-compatibility question directly: Snowflake, BigQuery, Airflow, dbt, Tableau, Looker, Kafka — with community contribution depth shown per integration.
User Benefit
Engineers get a personalized answer to "does it support my stack?" in 5 seconds — turning an unresolved question into a clear yes.
Business Benefit
Integration breadth is a competitive moat. The searchable catalog makes that moat visible and personally relevant rather than burying it in a logo grid that no one evaluates.
Measured Impact

From 3K+ deployments
to 11K+ GitHub stars

0
Bounce Rate Down
Reduction in homepage bounce rate within 90 days of redesign launch
0
More Sandbox Signups
3x increase in free sandbox trial activations from homepage traffic
0
Deeper Funnel Engagement
Increase in visitors reaching the sandbox trial or documentation pages — a proxy for intent, tracked alongside session duration to filter casual browsing from qualified exploration
0
Community Members
Active open-source community members across Slack, GitHub, and 430+ code contributors — supported by clearer community entry points on the site

Key Learnings

What this project taught me

01
Treat the marketing site as a product
The shift from "what does this page contain?" to "what does a data engineer need to believe by the time they leave this page?" changed everything. Once the site was treated as a product with users, jobs-to-be-done, and conversion funnels, every design decision had a clear evaluation criterion.
02
Visual language is the first credential for developer audiences
Developers evaluate trust before they evaluate features. A data-viz aesthetic signals domain expertise before a single sentence is read. For technical audiences, the visual language does the credential work that sales copy can't — because developers can identify authenticity on sight.
03
Multi-persona sites need architectural decisions, not compromise
Trying to design one homepage that works equally for developers, data engineers, and enterprise buyers produces a page that works for none of them. The resolution is layered hierarchy — each persona finds their signal at their natural depth — not averaging them into a single message that fails everyone.
04
Ship and iterate beats perfect pre-launch
The homepage hero copy was iterated 3 times in the first 60 days based on A/B results. The original launch was the hypothesis; the data told us where it was wrong. Instrumenting analytics from day one made that iteration possible — and the final version was better than anything we could have designed without real user behavior.
The long version Research, personas, flows
& the thinking underneath 6 sections · the reasoning behind every decision above
“

"The most useful reframe in this project was treating the marketing website as a product — with users, jobs-to-be-done, conversion funnels, and measurable outcomes. The specific thing that changed once we applied that lens: we stopped asking 'what does this page contain?' and started asking 'what does a data engineer need to believe by the time they leave this page?' Those are completely different questions, and they produce completely different pages."

Rupesh Chavan — Lead Product Designer