Tecton Design System
Tecton is a machine learning feature platform that enables teams to build, deploy, and manage production-ready ML features at scale. The platform serves ML engineers and data scientists who need to operationalize feature pipelines with speed and reliability.
Before I joined, the team had adopted an off-the-shelf design system from another company, but heavily customized it.
Role | Timeline |
---|---|
I led the strategy, design, and execution of the new design system.
| 6 months |
Business Need
Slow development cycles: The previous system slowed down feature development due to its complexity.
High design & engineering overhead: The customized design system required significant maintenance, eating into both design and development resources.
Scaling challenges: As the platform expanded, inconsistencies in UI patterns and interactions became more pronounced.
Key Objectives & Impact
Build a design system tailored for ML workflows (e.g., data-heavy UIs, complex filtering, and real-time monitoring).
Increase engineering efficiency—developers should be able to ship UI enhancements and fixes without needing a designer for every change.
Ensure consistency across the platform—users should be able to recognize patterns and workflows across different parts of the app.
Research & Process

Audit & Competitive Research
Conducted an audit of existing components, identifying gaps, inconsistencies, and areas of inefficiency.
Researched industry-standard design systems (e.g., Ant Design, Vercel’s Geist) to assess best practices for developer-heavy products.
Key Tradeoffs

Adopting an off-the-shelf UI kit vs. building from scratch
Stakeholder perspectives:
Engineers preferred fully custom components to optimize performance.
PMs wanted faster time-to-market, which an existing library could provide.
Resolution: We built foundational components from scratch but leveraged existing patterns for lower-priority UI elements.
User Flow & Use Cases
Mapped out key workflows (e.g., feature creation, data pipeline monitoring) to inform component prioritization.
Identified UI components that needed to be purpose-built for ML use cases rather than generic patterns.
Process
Developer Collaboration
Bi-weekly syncs with frontend engineering to review and refine implementation.
Averaged two components shipped per week, keeping momentum high.
Prioritizing MVP components
Stakeholder perspectives:
Engineers wanted immediate solutions for pressing UI issues.
Designers preferred a structured, comprehensive rollout to avoid inconsistencies.
Resolution: We launched an MVP set of core components first (tables, overlays, filtering, forms), then iterated on additional elements in subsequent quarters.
Adoption Strategy
Gradual rollout, starting with new features and migrating existing components over time.
Internal documentation and a Figma component library to drive adoption.
Slack channels & workshops to support designers and engineers in using the system effectively.
Design
Wireframes & Prototyping
Initial prototypes focused on simplifying interactions while ensuring powerful data manipulation capabilities.
Iterated based on feedback from internal teams and user testing.
Base Design System
Typography

When designing the interface for our machine learning product, clarity and usability were key. Given the complexity of machine learning workflows, we needed a typeface that could present dense data clearly while maintaining a modern aesthetic. Inter was the ideal choice due to its readability, versatility, and clean design.
Optimized for Complex Data
Inter’s precise letterforms and tuned spacing ensure excellent legibility, even when displaying detailed metrics and technical information. This makes it easier for users to quickly interpret complex outputs.
Color

Our color strategy needed to reflect our brand identity while supporting a highly functional interface. With red as our primary brand color, we faced the challenge of using it effectively without overwhelming the user. Additionally, dark mode was essential to align with developer preferences and improve usability in low-light environments.
Developer-Focused Dark Mode
Dark mode wasn’t just an aesthetic choice—it was a functional one. We created a thoughtful dark theme with high contrast and desaturated accents, reducing eye strain for developers working long hours while maintaining the product’s personality.
Flexible and Accessible
Every color was tested for WCAG compliance, ensuring accessibility across both light and dark themes. This allowed us to deliver a visually cohesive experience while meeting the needs of all users.
Iconography

Our iconography system needed to balance customization for unique product concepts with the speed of using a pre-existing library. I designed a series of custom icons to represent Tecton-specific ideas, ensuring clarity for complex workflows while reinforcing our product’s identity.
Custom Icons for Unique Concepts
Some concepts—like feature pipelines and real-time data flows—had no clear visual analog in standard libraries. I created a bespoke set of icons to represent these ideas, ensuring they were simple, scalable, and aligned with our visual language.
Efficiency with MyIcons
To speed up delivery, I adopted the MyIcons library for common actions and UI elements. This approach allowed us to maintain a consistent look while focusing custom design work on areas where unique visuals added the most value.
Component Library
Key components designed and implemented:
Tables

Tables were a core UI component for Tecton, essential for managing and exploring large datasets in machine learning workflows. I designed a flexible, scalable table system that handled complex data structures while maintaining performance and usability across varied use cases.
Select menus

Select menus were critical for navigating complex options in Tecton’s interface. I improved interaction patterns to make selection faster and more efficient, optimizing the experience for users handling large datasets and frequent inputs.
Usability Insights & Iterations
Error Messaging

Error handling was crucial for improving user trust in Tecton’s complex workflows. I standardized error messages to ensure they were clear, actionable, and consistent, helping the team identify issues faster and ship a more reliable product.
Showing more info by default

Early usability testing revealed that engineers preferred more transparency over simplified UIs—we adjusted designs to surface more system details rather than hide them.
Better filtering

Filters were essential for managing large datasets in Tecton. I expanded filtering options and optimized table interactions, making it faster and easier for users to find and act on specific data, especially when performing bulk operations.
Impact
Quantifiable Business & User Impact
60% of the product was using the new system within the first quarter.
All new features were designed using the design system.
35% faster design bug fixes, measured through engineering sprint velocity.
Qualitative Wins
Developers reported major improvements in velocity:
“Updating the search components was seamless. The built-in styling is polished and consistent, so everything we ship looks great without extra work.” – Front-End Engineering Manager
“This is amazing—everything is so well-organized. The styling is clean and flexible, which made it easy to build the latest feature and have it look perfect right out of the box.” – Front-End Engineer
Product & design teams benefited from clearer UI standards:
Less back-and-forth between design and development.
More consistent user experience across the platform.
Key Takeaways
Move fast, prioritize core needs. Instead of trying to perfect a full-fledged system, focus on high-impact components first.
Integrate into developer workflows early. Ensuring immediate adoption from engineering is key to long-term success.
Balance developer and user needs. Technical users prefer transparency over oversimplification—design accordingly.