Open Brain Platform (OBP) lets scientists build, simulate and publish digital brain models, from ion channels to single neurons, circuits and brain regions. I was its only product designer, product owner of the Virtual Lab and a frontend engineer, from EPFL's Blue Brain Project in 2022 to the Open Brain Institute's public launch in 2025.
Open Brain Platform (OBP) lets scientists build, simulate and publish digital brain models, from ion channels to single neurons, circuits and brain regions. I was its only product designer, product owner of the Virtual Lab and a frontend engineer, from EPFL's Blue Brain Project in 2022 to the Open Brain Institute's public launch in 2025.
When Blue Brain ended, the team went from 142 to 32 people and the product had to sustain the institute: 600 new virtual labs a year, plus grants and university partnerships. No product did the whole journey, and its users were experts with very different needs.
110 one-on-one interviews (86 researchers, 24 medical researchers), 30–34 workshops and co-creation sessions, and task-based tests with questionnaires and group debriefs. After launch, analytics told us who to call.
A platform organised around how experiments are actually run, where research decided what shipped: a 2D layer viewer instead of a 3D brain, view modes and a provenance map for circuits, and a validator that turned uploads into a guided task.
1,200+Virtual labs
Created in the 18 months since the March 2025 public launch, about 1.3× the pace of the 600-a-year target.
86% → 8%Circuit viewer bounce
After shipping view modes and the provenance map. Measured in Matomo.
2.1k → 30k+Entities uploaded
From single-entity upload to the multi-entity validator.
12Partner universities
Across Europe, the USA and China, for research and teaching.
All screens on this page use placeholder data: research on OBP belongs to the scientists and is confidential.
The only product designer on the team. Product owner of the Virtual Lab (members, projects, credits), Help & Feedback, and Reports (scientific publishing). Frontend engineer.
2022 – 2026. From EPFL's Blue Brain Project to the Open Brain Institute. Public launch in March 2025.
32 people (142 at the start): 18 scientists, 4 full-stack, 4 backend, 3 visualisation and 2 HPC engineers, and me. Management: a CEO, a CTO, a Chief Science Officer and 2 project managers.
01
600 new virtual labs a year, or no institute.
OBP started inside EPFL's Blue Brain Project as tooling for its own scientists. When the project ended, the Open Brain Institute took it over with 32 of the original 142 people and a new brief: the platform now had to sustain the institute. The targets were explicit: 600 new virtual labs a year, grants to push simulation to brain-region and whole-brain scale, and partnerships with universities. Nobody else covers the whole journey. The Allen Institute offers an atlas and explorers; other tools simulate one or two neurons. OBP had to take a scientist from exploring data to building, simulating and publishing, in one place.
The virtual lab home: three ways in (explore data, launch a workflow, start from a notebook) and an assistant that can drive the whole platform.
02
Four years of listening to scientists who know their field far better than any designer.
Research didn't stop at launch. Every release went through the same loop, and analytics told us who to talk to next.
110Interviews
One-on-one: 86 researchers and 24 medical researchers.
30–34Workshops
Topic workshops (model visualisation, simulation analysis) and co-creation sessions of 4 to 12 people.
Task-based tests
Users run real tasks (explore data, launch a workflow, build a neuron model) then fill in a questionnaire.
Group debrief
What was hard, where they got stuck, what they would change or add.
Ship & measure
Matomo shows where users drop off, and who to contact next.
Two primary personas, each anchored to a part of the product.
Build models, from ion channels to circuits. They need rigour, provenance and control over every parameter.
Work on Parkinson's, Alzheimer's, ALS and other neurodegenerative diseases. They need to simulate, not to model from scratch.
Neuroscience students and teachers, and AI teams using biological circuits to train models differently.
03
Three findings that overturned what we thought users wanted.
An interactive 3D brain would be the most engaging way into the data: a spatial, visual entry point to browse brain regions and everything recorded in them, from morphologies to electrophysiology.
Researchers found the 3D view inaccurate, and it said nothing about how brain regions connect. They asked for a detailed 2D view showing relationships with surrounding areas.
A layer-by-layer 2D viewer, used by default, that shows the type of connection with adjacent regions.
More traffic in Explore, especially Data. Thank-you notes from scientists. A partnership with a company producing accurate biological brain data to enrich the 2D views. And new requests, like dendrograms.
A circuit holds a massive amount of information: synapse concentration, bouton density, layer thickness and more. The viewer showed it at one of two extremes. Either everything at once, which was unreadable, or so little that nothing could be exploited.
Users left the circuit viewer: 86% bounce. Many took screenshots and wrote to support to ask what they were looking at, or to say nothing was usable.
View modes instead of filters. Each mode answers one question (synapse concentration, bouton density, layer thickness), and the same circuit can be read as a table, in 2D or in 3D.
Together with the provenance map, bounce on the circuit viewer fell from 86% to 8%. Usability stopped coming up in workshops.
Each entity, such as a single neuron model, only linked to the source it was created or extracted from. To know its full story (where it came from, and where it was reused), scientists had to follow those links one by one and rebuild the chain themselves.
Tracing where a neuron model came from, and where it was reused, was painful. Scientists need exactly that to decide what is worth extracting and citing in a paper.
An interactive relationship map with advanced filters, to trace any entity's origins and reuse across the platform.
Scientists can judge what is worth viewing, extracting and using in publications. With the view modes, it brought circuit viewer bounce from 86% to 8%.
86% → 8%Circuit viewer bounce rate
Before: scientists screenshotting the viewer and writing to support to ask what they were looking at. After: view modes and the provenance map.
Measured in Matomo. The features were released and presented at a public workshop with scientists, two weeks after full implementation.
04
Choosing what to build, and what not to.
Workflows are the core of OBP. With scientists and the executive board, I ran workshops to organise everything the platform does into a few verbs (Build, Simulate, Extract, Optimize, Validate), then tested three ways into them.
Pick a category, then a type (ion channel, synaptome, circuit…), then reuse an existing workflow or create a new one.
Too complex, and it reversed the natural order in which scientists run an experiment.
The activity table lives on as a way to browse and explore existing workflows.
PrincipleThe interface follows the order in which scientists actually run an experiment.
Every workflow shares one structure: a guided setup on the left, and the context needed to decide on the right.
Entities were uploaded one at a time, and each had to match OBP's schema and taxonomy exactly: many fields, strict conventions, typed data. Preparing a single upload was never quick, let alone a lab's whole dataset.
Uploads must match OBP's schema and taxonomy: many fields, strict conventions, typed data. We could not silently replace scientists' data with our own. They had to fix it themselves, and understand why.
The validator: a spreadsheet-like table that flags errors and warnings per cell, row and column, then imports a clean batch. It took several rounds to make it simple and fast. Batch edits were the hardest part, because one change could trigger errors elsewhere.
Uploaded entities went from 2,100 with single upload to 30,000+ with multi-entity upload. When analytics showed users importing batches of 100, we contacted them and simplified the flow with their feedback. Every month, 10 to 20 more users upload 50+ entities.
I built the roadmap with the CTO and scientific leads, and every Friday I defended my tickets and priorities using RICE and impact/effort. Some of the most important decisions were refusals.
Scientists wanted to showcase their best work. But those cases were so complex they hid how to get started with the tools. The home page has to get a new lab going.
A virtual lab is where projects start, not where work happens. Admins invite researchers and manage the budget; members join projects.
PrincipleA virtual lab is an administrative space. Projects are where science happens.
People think in dollars, euros or francs. The hard part was knowing what an OBP credit actually buys, and whether it pays for compute or storage.
Cost transparency everywhere, not only at launch: a breakdown before simulations run and, as I pushed for, live examples of what an amount of credits can achieve as you type it. Lab owners buy credits and allocate budgets to projects, like a real lab.
Jupyter notebooks are the standard in science. Asking researchers to learn OBP's architecture, restructure their experiment and run it again, with new chances of error, was a barrier to adoption. So notebooks run natively inside OBP: scientists start in a familiar environment, and I proposed tips inside their notebooks pointing to the native OBP tool that does the same step. Ready-made notebooks also show what the platform can do.
05
I designed it, then built a large part of it.
Code was part of my design process: I prototyped the circuit atlas and Explore as proofs of concept, and shipped the parts I owned end to end.
Virtual lab home and project navigator Alerts and system messages Help and feedback Reports, where scientists gather the models and simulations they generated into a full scientific paper: peer review with comments, feedback and isolated discussions, a draft-to-publish flow, and embedded simulation viewers and entity details such as neuron morphologies and electrophysiology
openbraininstitute.org, built end to end.
Early proofs of concept for the circuit atlas and Explore.
OBI Assistant can drive the whole platform: navigate, build models, launch simulations, open notebooks. Built by Jan, Nicolas and Boris from the ML team; I designed and built its interface.
06
What the work added up to.
They explore data they can trust: an accurate 2D view by default, view modes that answer one question at a time, and the full provenance of every model. Thank-you notes followed, and usability stopped coming up in workshops.
They fix their own data before import, in a spreadsheet-like validator, instead of uploading one entity at a time: from 2,100 to 30,000+ entities uploaded.
They see what a simulation costs before it runs, and allocate budgets to projects like a real lab.
It needed 600 new virtual labs a year to exist. Scientists created 1,200+ in 18 months, with 12 partner universities.
1,200+Virtual labs
In 18 months, from zero.
12Partner universities
Europe, USA and China.
51People in the largest lab
Labs grow by inviting colleagues.
30k+Entities uploaded
Up from 2,100.
After the Explore and circuit changes, workshops only raised aesthetic details. Usability stopped coming up.
A company producing accurate biological brain data partnered with OBI to enrich the 2D views.
Thank-you notes for the 2D default, and users asking for more, like dendrograms.
I would design onboarding backwards, starting from the outcome. OBP is huge, with many tools and features, and newcomers see the steps long before they see the point. Each workflow would open with a conversation with the OBI Assistant around a real publication: here is the result, here is the figure. From there, users could go back in time and understand what led to each result and artifact before building their own.
