Pipelines
The Product
Code Ocean’s mission is to make computational science reproducible and collaborative. Scientists use “Capsules”; self-contained environments that ensure experiments can be re-run exactly as before. As data volumes grew, scientists needed a way to automate multi-step analyses, from preprocessing to validation, without writing hundreds of lines of code.
The goal was to create a visual, reproducible workflow builder that unites Capsules and Data into Pipelines, enabling both computational and bench scientists to automate confidently.
My Role
As Lead Product Designer, I shepherded the feature from concept through release.
My role spanned research, system mapping, UX/UI design, and hands-on prototyping, working closely with engineers to ensure that what looked simple also performed flawlessly under heavy computational loads.
From the first workshop, I made collaboration the backbone of the process.
Scientists sketched their ideal workflows on whiteboards; engineers explained what was technically feasible; together we iterated on a design language that spoke both human and machine.
Our feedback loop ran weekly: ideas → prototype → test → refine.
The Challenge
At the heart of the problem was complexity visualization, representing dozens of interdependent steps without overwhelming the user.
We also needed to handle performance for pipelines exceeding 100 nodes, display real-time execution feedback, and surface lineage linking every dataset and environment without modifing the extisting IDE strcutre.
“It takes a lot of time and effort to build pipelines direclty with code, it needs imagination and memory,”
— Computational Biologist, Bioinformatics Team
The Research
Findings
Early interviews with computational and bench scientists exposed a universal pain point: “I lose track of what step triggered what.”
Key findings:
Bench scientists feared breaking scripts, lacked confidence, and needed guided workflows.
Computational scientists demanded transparency, debug visibility, and per-step resource control.
Most tools lacked contextual feedback, users had to open terminals to see progress or errors.
Quantitative patterns:
70 % of users had failed to reproduce an analysis within 6 months.
50 % reported unclear pipeline dependencies.
Average onboarding time: 1–2 weeks for non-coders.
Takeaways
Only Terra combined visual simplicity and per-step compute control and was positioned as the leader.
Nextflow Tower excelled in performance but lacked accessibility.
Galaxy was friendly but not scalable.
“If I could reuse a trusted pipeline with my data, that would save me weeks.”
— Computational Biologist, Bioinformatics Team
The Design
Design Principles
The solution evolved around three core principles: progressive disclosure, transparency, and composability.
We introduced a node-based canvas where each operation alignment, QC, analysis appeared as a draggable card.
Beginners could rely on curated templates; advanced users could open a code view for fine control.
A parameter drawer offered contextual help, defaults, and validation so users never felt lost.
During execution, a run timeline revealed logs, statuses, and compute usage in real time.
Finally, a lineage panel stitched together data, capsules, and results, making reproducibility visible, not abstract.
Architecture
Graph canvas (core view).
Layer controls.
Data and Capusles File Tree.
Visualization Toggle: code vs GUI
Run and versions Timeline.
Asset Drag-n-Drop.
Configuration modal for connections and capsules.
Import Pipelines from nf-core modal.
Share with other teams capabilities.
Solution Overview
The final Pipelines design turned scripting complexity into a clear, visual experience.
Users can now see every step, dataset, and dependency directly in a graph view, with real-time visibility through the Run Timeline on the right and seamless file access on the left.
This single view unified creation, monitoring, and reproducibility, eliminating context-switching between code, data, and execution logs.
As a result, scientists build and debug pipelines faster, with full confidence in every run.
Validation & Outcomes
Testing & Validation
Over three testing cycles we watched scientists build, break, and rebuild pipelines in prototype environments. Each round exposed subtle friction, terms that confused users, visual overload in large graphs, or missing feedback during runs.
By the final iteration, the flow felt natural: scientists could create working pipelines in minutes, not hours.
Quantitatively, setup time dropped 40 %, first-run success more than doubled, and satisfaction rose +1.6 points on our five-point scale.
“I can now understand what’s happening while my code and parameters are running.”
— Computational Scientist, internal pilot
Reflection
Pipelines reaffirmed that progressive disclosure is the bridge between accessibility and power.
Scientists don’t reject complexity, they reject confusion and by revealing only what’s needed, when it’s needed, we earned trust without dumbing things down.
Our next frontier is collaboration and intelligence:
multi-user editing, AI-driven error detection, and a library of public workflow templates that democratize automation even further.
Designing Pipelines taught me how to translate deep technical logic into intuitive visual language. It reminded me that design’s highest calling is not decoration, but understanding, building bridges between expertise levels so everyone can create with confidence.
This project remains one of the clearest demonstrations of how thoughtful UX can turn scientific rigor into everyday usability.
Thankyou :)
"Good design is obvious. Great design is transparent."
- Joe Sparano
