B2B
B2B
B2B
Web App
Web App
Web App
INTEL X
INTEL X
INTEL X
MachineLab is an end-to-end ML platform concept designed to
MachineLab is an end-to-end ML platform concept designed to
take models from dataset setup to deployment and monitoring in one guided workflow. It helps teams ship and improve prediction models faster by reducing tool-switching and making progress visible.
take models from dataset setup to deployment and monitoring in one guided workflow. It helps teams ship and improve prediction models faster by reducing tool-switching and making progress visible.
MachineLab is an end-to-end ML platform concept designed to take models from dataset setup to deployment and monitoring in one guided workflow. It helps teams ship and improve prediction models faster by reducing tool-switching and making progress visible.
Challenge
Challenge
Challenge
Ascendeum’s ML work was slowed by fragmented tools and handoffs across data prep, training, deployment, and monitoring, creating delays in shipping models that impact ad performance.
Teams spent disproportionate time coordinating workflows instead of improving models that drive client ROI.
Ascendeum’s ML work was slowed by fragmented tools and handoffs across data prep, training, deployment, and monitoring, creating delays in shipping models that impact ad performance.
Teams spent disproportionate time coordinating workflows instead of improving models that drive client ROI.
Ascendeum’s ML work was slowed by fragmented tools and handoffs across data prep, training, deployment, and monitoring, creating delays in shipping models that impact ad performance.
Teams spent disproportionate time coordinating workflows instead of improving models that drive client ROI.
Design Approach
Design Approach
Design Approach
I used a research-led design approach—interviewing Ascendeum data scientists/admins and shadowing real ML workflows—to map a consistent AdTech model lifecycle and identify where time was lost.
We then designed a stage-based “Model Journey” where models move as cards through each step with progressive disclosure, making the flow beginner-friendly while keeping expert controls for tuning, deployment, and monitoring.
I used a research-led design approach—interviewing Ascendeum data scientists/admins and shadowing real ML workflows—to map a consistent AdTech model lifecycle and identify where time was lost.
We then designed a stage-based “Model Journey” where models move as cards through each step with progressive disclosure, making the flow beginner-friendly while keeping expert controls for tuning, deployment, and monitoring.
I used a research-led design approach—interviewing Ascendeum data scientists/admins and shadowing real ML workflows—to map a consistent AdTech model lifecycle and identify where time was lost.
We then designed a stage-based “Model Journey” where models move as cards through each step with progressive disclosure, making the flow beginner-friendly while keeping expert controls for tuning, deployment, and monitoring.
Space
Space
Space
Machine Learning
Machine Learning
Machine Learning
Role
Role
Product Designer
Product Designer
Product Designer
Users
Users
Users
Data Scientists
Data Scientists
Data Scientists
Team
Team
Team
2 Designers
2 Designers
2 Designers
How we started?
How we started?
How we started?
Conducted structured user interviews across skill levels (beginners, advanced ML engineers) plus platform/admin stakeholders to understand AdTech workflows and constraints.
Shadowed end-to-end model workflows (data, train, deploy, monitor) to capture where time is lost: tool switching, repeated setup, unclear ownership, and debugging production issues.
Conducted structured user interviews across skill levels (beginners, advanced ML engineers) plus platform/admin stakeholders to understand AdTech workflows and constraints.
Shadowed end-to-end model workflows (data, train, deploy, monitor) to capture where time is lost: tool switching, repeated setup, unclear ownership, and debugging production issues.
Conducted structured user interviews across skill levels (beginners, advanced ML engineers) plus platform/admin stakeholders to understand AdTech workflows and constraints.
Shadowed end-to-end model workflows (data, train, deploy, monitor) to capture where time is lost: tool switching, repeated setup, unclear ownership, and debugging production issues.
Critical Pain Points
Critical Pain Points
Critical Pain Points
Efficiency bottlenecks and lack of advanced workflow support.
Visibility and governance challenges across team workflows.
Efficiency bottlenecks and lack of advanced workflow support.
Visibility and governance challenges across team workflows.
Efficiency bottlenecks and lack of advanced workflow support.
Visibility and governance challenges across team workflows.
4-6 hrs/week wasted on data movement.
4-6 hrs/week wasted on data movement.
4-6 hrs/week wasted on data movement.
45–90 minutes spent just recreating setup (datasets, splits, feature steps, metrics) before training.
45–90 minutes spent just recreating setup (datasets, splits, feature steps, metrics) before training.
45–90 minutes spent just recreating setup (datasets, splits, feature steps, metrics) before training.
Key Research Insights
Key Research Insights
Key Research Insights
Need for stage-based workflow that mirrors natural ML progression.
Progressive disclosure strategy essential for serving multiple skill levels.
Built-in sharing, versioning, and knowledge transfer capabilities needed
Need for stage-based workflow that mirrors natural ML progression.
Progressive disclosure strategy essential for serving multiple skill levels.
Built-in sharing, versioning, and knowledge transfer capabilities needed
Need for stage-based workflow that mirrors natural ML progression.
Progressive disclosure strategy essential for serving multiple skill levels.
Built-in sharing, versioning, and knowledge transfer capabilities needed
Swimlane / Kanban structure easily accommodates the required complexity.
Swimlane / Kanban structure easily accommodates the required complexity.
Swimlane / Kanban structure easily accommodates the required complexity.
Stage-based visual pipeline with drag-and-drop model cards
Stage-based visual pipeline with drag-and-drop model cards
Progressive disclosure, guided experience, version transparency
Progressive disclosure, guided experience, version transparency
Drag model cards through stages
Drag model cards through stages
Stage Based Visual Pipeline
Stage Based Visual Pipeline
Progressive Disclosure
Progressive Disclosure
Progressive Disclosure
Progressive Disclosure
Viewing Data in the Application
Viewing Data in the Application
Drag and Drop Model Cards
Drag and Drop Model Cards
Impact
Impact
Impact
Faster iteration loops: reduced coordination overhead so models and updates can move from experimentation to production faster.
Better onboarding: new team members can follow a consistent step-by-step workflow rather than learning many disconnected tools.
Experiment setup time reduced from 45-90 min to 27-45min per run with a unified workflow and reusable configuration
Data movement overhead reduced from 4-6 hrs/week to less than 3-4hrs/week by reducing rework and repeated transfers
Faster iteration loops: reduced coordination overhead so models and updates can move from experimentation to production faster.
Better onboarding: new team members can follow a consistent step-by-step workflow rather than learning many disconnected tools.
Experiment setup time reduced from 45-90 min to 27-45min per run with a unified workflow and reusable configuration
Data movement overhead reduced from 4-6 hrs/week to less than 3-4hrs/week by reducing rework and repeated transfers
Faster iteration loops: reduced coordination overhead so models and updates can move from experimentation to production faster.
Better onboarding: new team members can follow a consistent step-by-step workflow rather than learning many disconnected tools.
Experiment setup time reduced from 45-90 min to 27-45min per run with a unified workflow and reusable configuration
Data movement overhead reduced from 4-6 hrs/week to less than 3-4hrs/week by reducing rework and repeated transfers
Learnings
Learnings
Learnings
Stage-based workflows reduce cognitive load because they match how teams naturally think: “what step are we in?”.
Progressive disclosure is essential for mixed-seniority teams—one UI can serve beginners and experts without overwhelming either.
Stage-based workflows reduce cognitive load because they match how teams naturally think: “what step are we in?”.
Progressive disclosure is essential for mixed-seniority teams—one UI can serve beginners and experts without overwhelming either.
Stage-based workflows reduce cognitive load because they match how teams naturally think: “what step are we in?”.
Progressive disclosure is essential for mixed-seniority teams—one UI can serve beginners and experts without overwhelming either.