Summary
Nexus Fusion gave neuroscientists at EPFL's Blue Brain Project access to millions of research resources. As Product Design Lead and frontend engineer, I refocused an engineering-led product on how researchers actually find and use data.
Nexus Fusion
My role in the project
2014_2024
Summary
Nexus Fusion gave neuroscientists at EPFL's Blue Brain Project access to millions of research resources. As Product Design Lead and frontend engineer, I refocused an engineering-led product on how researchers actually find and use data.
Challenge
Access went mostly through an SDK and the tables it generated, with no overall view. And researchers used to the terminal were not to be forced into another interface.
Research
Interviews with scientists from several fields of neuroscience, scenario-based user tests, and usage statistics I requested from three backend engineers: 86% of the data had not been consulted after publication.
Solution
A web application bringing publication management, 3D visualisation and documentation together, and a relationship module I took from a Figma prototype to an approved D3.js proof of concept.
Key figures
86%Data never consulted
Share of data not opened after it was put online, on the measured scope. A diagnostic I initiated, not an outcome.
100Real entities
Downloaded to build an isolated sandbox prototype of the relationship module.
20Neurons
In the test circuit that turned a contested proposal into an approved proof of concept.
4Team leads convinced
My manager and the heads of the NISE, computational neuroinformatics and HPC teams.
All screens use placeholder or demo data; research data stays confidential.
My role & context
- Role
Product Design Lead and frontend engineer. I led research and product design on Nexus Fusion.
- Product design
- UX research
- Product strategy
- Frontend
- Timeline
2022 – 2024 · Blue Brain Project, EPFL.
- Team
The NISE team in charge of Fusion, three backend engineers, and the computational neuroinformatics and HPC teams.
- Methods
- Interviews with scientists
- Usage statistics
- Scenario-based user tests
- Workshops
- Figma prototyping
- Sandbox prototype
- Stack
- Figma
- D3.js
01
The stakes
Millions of resources, and almost no way to find them.
An engineering-led platform
Nexus Fusion gave researchers access to millions of neuroscience resources. After several years of development led mainly by engineering, access went largely through an SDK and the tables it generated, with little overall view. Discovery and navigation were so heavy that researchers could not reach the pieces of data they needed inside each entity. One constraint shaped the work: the manager of the NISE team, in charge of Fusion, did not want researchers used to the terminal to be forced into another interface.
_05
The entry point: connect, browse studios, read the documentation.
Before the redesign
Access went through the SDK and generated tables.
02
Research
Talking to scientists, and measuring what nobody had measured.
How I built the diagnostic
Interviews
With scientists from different fields of neuroscience, to find the real needs and the problems of the existing version.
Usage statistics
I asked three backend engineers for what nobody was tracking: data uploaded, entities used in models or simulations, and how often data was viewed (never, by 1–10 users, by 11–20…).
Scenario-based tests
Researchers worked through realistic tasks from written scenarios, in test sessions and workshops.
Documented findings
Each issue recorded with the user's task, raw feedback, a recording and a design proposal.
Research in numbers
162+Scientists interviewed
5Fields of neuroscience
Biological experimentalists, simulation, cell, circuit and ion channel.
400+Test sessions
Plus 34 workshops.
Research artifacts
What the statistics showed
03
What research changed
Volume was not the problem. Visibility was.
Insight 01 — Discovery
86% of the data was never opened after it was published
Before
Resources were uploaded at scale, but mostly reachable through SDK-generated tables. Nobody knew how much of it was actually used.
Research showed
The statistics I requested showed that 86% of the data had not been consulted after being put online, on the measured scope. Researchers also could not reach the useful pieces of data inside entities: navigation and discovery were too massive.
We built
We brought the features together in a web application: publication management through studios, interactive 3D visualisation, documentation, and a more spaced-out update strategy.
Two ways to read a table
Table view structure: a standard mode, an advanced mode, filtering, and detail panes that expand on demand.
04
Product decisions
Turning a contested proposal into an approved experiment.
The relationship module, step by step
Entities were shown in isolation, yet science links them: a neuron belongs to a circuit, which belongs to a larger circuit. I proposed a module to explore resources through those relationships.
12Figma prototypes
A first interactive version of the idea.
100Sandbox prototype
Real entities downloaded to build an isolated prototype on actual data.
4Review
Presented to my manager and the heads of the NISE, computational neuroinformatics and HPC teams.
20Feasibility test
A 20-neuron circuit, proposed by the HPC team manager, to test the idea at small scale.
4Proofs of concept
Approved, built with D3.js on valid data provided by the head of computational neuroinformatics.
From Figma to proof of concept
The objections, and how they were answered
- Cost
The leads estimated the development cost of extracting that much data, connecting it and displaying it in real time.
- Priority
The NISE team manager opposed adding a complex tool while experiment reports were still in development.
- Data quality
The head of computational neuroinformatics questioned feasibility: some circuits have no clear, complete list of their neurons and synapses.
- Answer
The HPC team manager proposed a test on a small 20-neuron circuit. It showed the potential, with clear room to improve performance and data quality, and I was allowed to build a proof of concept.
05
Design engineering
Making a proposal concrete enough for engineers to judge it.
What I built
- Prototypes
A Figma prototype, then an isolated sandbox prototype built on 100 real entities.
- Proof of concept
The relationship module with D3.js, on valid data from the computational neuroinformatics team.
- Stack
- Figma
- D3.js
06
Outcome
An approved experiment, not yet a shipped feature.
Who it helped
- Researchers
They find and reuse data instead of losing it in SDK tables: data consulted went from 14% to 94%, with 2,600+ monthly active researchers on the web app.
- Data producers
Their work gets seen and reused: 12,000+ reports, 20,000+ experimental data and 150+ studios published.
- Engineering teams
A contested idea became a proof of concept they could judge on real data, approved by four team leads.
What changed after launch
14% → 94%Data consulted
Share of data opened after being put online, on the same scope as the initial 86% never consulted.
2,600+Monthly active researchers
On the Fusion web application.
12,000+Reports published
20,000+Experimental data published
150+Studios published
Where it stands
- Established
A usage diagnostic nobody had run before, and a contested proposal taken from Figma to an approved proof of concept on real data.
- Still open
The relationship module's feasibility at the scale of millions of resources, and its production release.
