Client
Kaskada is a Feature Engineering Studio designed to help data scientists and ML engineers create, transform, and manage features efficiently for machine learning models. It enables seamless feature creation, transformation, and storage, supporting both real-time and batch processing. With built-in governance, data lineage tracking, and operationalization, Kaskada ensures reproducibility and scalability while integrating with feature stores for consistent training and inference. Through the designs, research, and development of the MVP, Kaskada successfully secured over $10 million in Series A funding and established product-market fit, proving its value in streamlining feature engineering workflows and accelerating model development for AI-driven applications.
​In January 2023, DataStax, a real-time AI company, acquired Kaskada to enhance its capabilities in delivering instant, actionable insights for AI applications. This acquisition aimed to integrate Kaskada's technology into DataStax's portfolio, enabling organizations to build applications infused with real-time AI. Following the acquisition, DataStax announced plans to open-source Kaskada's core technology and develop a new machine-learning cloud service.
Overview
CLIENT | Kaskada
ROLE | Head of Product Design
PLATFORM | Web
PRODUCT| Beta - MVP
GOAL | Series-A Funding
TIMELINE | 6 Months
Responsibilities
IA - Structure, Application Mapping
UX - Competitive Analysis, User Research & User Testing
UI - Wireframing, Prototyping & High Fidelity Mockups
Product - Market Analysis, Marketing Website Development & Product Roadmap Creation
Problem History
Designing features is an art. Deploying features is a pain.​
Historically, feature engineering was a fragmented process due to siloed workstreams within data teams. Data engineers, ML engineers, and operations teams worked independently, often leading to feature inconsistencies, redundant efforts, and slow iteration cycles. Without a centralized system, teams manually built features using custom SQL queries, Python scripts, or Spark jobs, making it difficult to ensure consistency between training and inference environments. This lack of coordination resulted in data drift, brittle production pipelines, and inefficient compute usage, ultimately slowing down the deployment of machine learning models.
The shift to Feature Stores and Feature Engineering Studios (like Kaskada) modernized this approach by unifying feature workflows, enabling collaboration, and ensuring real-time feature availability. These platforms eliminated redundant processing, improved governance with feature versioning and lineage tracking, and allowed teams to reuse features efficiently across models. By bridging the gap between training and production, modern feature engineering solutions have accelerated model deployment, enhanced reliability, and unlocked scalable, real-time AI applications.
Problems to Solve
How do we align with the current
patterns that data scientists expect in
their software while providing new
experiences and enhanced workflow
options they didn't realize they needed?
01
How do we inspire the creative process
for data scientists and allow them to feel
empowered all while learning a new
tool and language?
02
How do we break the mold of
traditional siloed work streams and
encourage collaboration?
03
Product & UX
Old Customer Journey
Current data management teams are siloed into separate roles and different software. Many other data sources can feed a data scientist's feature engineering workflow and they are many times dependent on data engineers for clean data. This can take many months just for a data scientist to get the data they need to start their work, feature engineering was a fragmented process due to siloed workstreams within data teams.
Product & UX
Target Audience
Information Architecture
Familiarity Through Navigation
To design an intuitive Information Architecture (IA) for Kaskada, I conducted in-depth research using industry-standard tools like Jupyter, Adobe Analytics, Tableau, and Orange to understand how users navigate and interact with data-driven interfaces. These platforms each provided valuable insights:
-
Jupyter: Users expect a left-to-right flow when working with data, mirroring the way code cells execute sequentially in notebooks. This informed the structure of data ingestion, transformation, and output visualization within the studio.
-
Adobe Analytics & Tableau: These analytics tools emphasize visual hierarchy and modular workspaces, where users can drag, drop, and manipulate data elements dynamically. This reinforced the decision to create a central canvas for users to explore and iterate on feature engineering tasks.
-
Orange: As a visual programming and ML workflow tool, Orange provided insights into how users construct data pipelines, reinforcing the need for an IA that supports modular, reusable components for building and testing features.
Building on these navigation expectations, I structured the framework of the Information Architecture to support a left-to-right data flow, where users could seamlessly move from data selection → feature transformation → model integration → deployment. The central canvas served as an interactive workspace for real-time feature exploration, visualization, and iteration, ensuring a smooth user experience while maintaining flexibility for both exploratory data analysis and production feature engineering. By aligning with familiar UX patterns from these tools, the IA reduced friction for users transitioning from other data environments, enhancing usability and adoption.
UX/UI
The Job of Quality Schemas
In the Jobs-to-Be-Done (JTBD) framework, schemas in feature engineering serve the essential job of structuring and standardizing data to ensure consistency, usability, and efficiency in machine learning workflows.
A UI-driven visual schema can significantly enhance a data scientist’s feature engineering workflow by making schema definition, validation, and management more intuitive, interactive, and efficient.
-
Simplifies Schema Definition and Understanding – Instead of manually writing schema definitions, a visual representation allows data scientists to drag-and-drop data fields, specify data types, and set constraints in an intuitive way, reducing complexity.
-
Enhancing Feature Reusability – Visual schemas make it easier to organize and categorize features, allowing data scientists to reuse validated feature sets across models. This saves time and reduces redundant work by providing a clear understanding of the feature's purpose and structure.
-
Improving Collaboration – By visualizing schemas, multiple team members (data scientists, engineers, analysts) can collaborate more easily. A shared visual representation of the schema allows everyone to understand and contribute to the feature engineering process, promoting better communication and reducing the chance of duplication or misunderstandings.
​
UX/UI
Complex Code to Clean Clickable UI
Simplifying the data schema from multi-line code into a folder structure allowed for the UI to consume complex data schemas all while retaining a clean interface.
Create Project
Explore your data
Easy export
Create Project
Product & UX
New Customer Journey
The Kaskada platform connected all consumers of data pipelines, resulting in cleaner, more accurate, and faster feature engineering. By introducing a UI-driven visual schema, data scientists could now define, validate, and manage schemas interactively, reducing complexity and minimizing errors. This transformation allowed users to quickly explore feature dependencies, track lineage, and enforce data governance in real-time, eliminating the need for manual schema definitions and reducing the risk of inconsistencies.
What once took months of back-and-forth between teams was now reduced to days, as the platform’s intuitive UI facilitated rapid iteration, cross-team collaboration, and seamless deployment, ultimately accelerating the journey from raw data to production-ready features.
