Data Fusion & KMS

01 · THE PLATFORM

See it in action

This is EY's official presentation of Data Fusion. It walks through the main modules of the platform: data preparation, governance, data quality, discovery and operations.

Play video · 6:07

02 · CONTEXT

What is Data Fusion?

It is a cloud-native, AI-powered, multi-tenant platform built on Databricks. One platform serves many clients, each in its own isolated space, and a low-code / no-code interface lets business and technical teams work with data without writing code.

What the platform does

• 01

  Data Preparation

  Upload files, profile them to find patterns and anomalies, onboard new data sources and map raw data into curated datasets.

• 02

  Data Governance

  Define and enforce each client's rules, and follow the results in Data Quality Insights.

• 03

  Data Discovery

  Explore raw and curated data with SQL queries in the Data Explorer.

• 04

  Data Operations

  Track the status and execution history of every data job.

• 05

  Configurations

  Query the platform's metadata through the Metadata Explorer.

• 06

  App Gallery

  AI and machine learning apps that extend the platform, from anomaly detection to document intelligence.

AI built into the workflow

• AI

  Data Mapping Agent

  Creates or refines a data mapping from a prompt written in natural language.

• AI

  Rule Agent

  Writes the SQL of a data quality rule automatically, and the user can still edit it.

• AI

  Sensitive data detection

  Finds personal and sensitive information inside the data.

• AI

  AI-driven data quality

  Spots anomalies and outliers before they turn into problems.

Screens from the video

03 · MY ROLE

Lead designer of the platform

I make the design decisions for Data Fusion and I execute them: from understanding a requirement with engineering, to the flow, to the final high-fidelity screens that go into development. I also design the variations of the platform, like KMS and the micro frontend apps that are joining it.

• 01

  Design decisions

  I define how each feature works and looks, and I defend those decisions with stakeholders and tech leads.

• 02

  End-to-end execution

  Flows, wireframes, prototypes and high-fidelity screens, built with Motif, EY's design system.

• 03

  Translating complexity

  I turn data engineering requirements into interfaces that business users can understand and use.

• 04

  Quality after handoff

  I review what was built against the design and flag gaps before they reach production.

Who I work with every day

One platform, several products

• Data Fusion

  The core platform: ingestion, governance, data quality, discovery and operations.

• KMS Fusion

  A new app built on top of Data Fusion that turns unstructured documents into a searchable knowledge library.

• Micro frontend apps

  A navigation model that lets independent apps live inside Data Fusion as one single experience.

04 · THE CHALLENGE

Learning the language of data

My biggest challenge was not the interface, it was the domain. Data engineering is full of concepts that are invisible to most people, and I could not design good screens for something I did not understand.

How I closed the gap

1. 01

   Asked a lot of questions

   Daily contact with data engineers and tech leads. I kept asking until I could explain each concept in my own words.

2. 02

   Desk research

   I studied ETL, data quality, governance, ontologies and the difference between structured and unstructured data.

3. 03

   Wrote it down

   Guide documents and flowcharts explaining the key concepts, validated with stakeholders. They became a reference for the team too.

4. 04

   Designed for the non-expert

   Having been the beginner, I knew where the language was too technical, and pushed for clearer labels and guided flows.

A few of the words I had to learn

ETL

Extract, Transform, Load. Taking data from a source, reshaping it and storing it where it is useful.

Data profiling

Scanning a dataset to understand it: its structure, value distribution, empty fields and outliers.

Data mapping

Defining how the fields of a raw source become fields of a curated, standard model.

Data quality rule

A check the data must pass, like "every account has an owner". Usually written in SQL.

PII / PCI

Personal and payment card information. Sensitive data that must be found and protected.

Tenant

One client's isolated space inside a platform shared by many clients.

Ontology

A map of concepts and how they relate to each other, like "Lasagna contains Cheese".

Unstructured data

Information that does not live in tables, like PDFs, forms and reports.