

Amazon Q Data Science
Build machine learning models and solve business problems using natural language
THE CHALLENGE
Our data engineers spent hours joining and preparing the data to remove anomalies. This manual process significantly slowed down our ability to deliver timely insights for data-driven decisions.
Chief Data Officer
Financial media and technology company
Citizen data scientists often struggle to keep pace with the rapid evolution of AI and machine learning, facing challenges in deciding between using foundation models or building custom solutions. They lack deep expertise in translating business problems into machine learning (ML) tasks and in understanding the full ML lifecycle — from data preparation and feature engineering to model development. Additionally, they need greater control and transparency in ML workflows to trace data origins, mitigate bias, and troubleshoot errors effectively.
ROLE
Lead UX designer
DURATION
5 months (June 2024 - November 2024)
KEY RESPONSIBILITIES
Participated in customer sessions to understand key pain points and translated findings into product requirements
Created end-to-end prototypes and gathered feedback from users and customers
Conducted usability tests and refined designs based on feedback
Partnered with PM, Engineering, and Science teams to prioritize MLP requirements and inform future roadmap
TEAM
UX: 3 (1 UXD, 1 UXR, 1 UXM)
PM: 3
Engineering: 8 (2 SDMs, 3 FEs, 3 BEs)
Science: 6
DESIGN PROCESS
Ideation
My design process began with a cross-functional ideation workshop with PM, Engineering, and Science to surface customer pain points. The opportunities we identified didn't just inform the design, they shaped the roadmap and organizational strategy.

Research
During the research phase, I conducted a competitive analysis of comparable products and participated in three customer sessions, capturing insights in a centralized document. Throughout research and iteration, I also ran usability tests with data scientists to gather feedback and refine the designs accordingly.

Concept generation
Throughout the research and concept generation stages, I created multiple concept variations based on initial customer pain points, gathered feedback, and iterated accordingly.

Concept evaluation
For the usability testing sessions, I built an end-to-end prototype that incorporated prior customer feedback to validate the full user journey with data scientist users.

key design decisions
Getting started
I want to see a broader variety of machine learning problem types the tool can solve so I can get a better understanding of the capabilities.
Senior PM, Base Pricing
The redesigned landing page features a hero section that clearly communicates the AI agent’s value proposition and includes a call-to-action to guide citizen data scientists toward their desired outcomes.

The AI agent landing page provides a curated set of example prompts, each tailored to business outcomes, enabling citizen data scientists to experiment based on their role, industry, or specific use case.

Chat interactions
It'll be more difficult for me to trust the system when I don't understand what it’s doing.
Solution Architect, ML Services
During chat interactions, the AI agent offers follow‑up prompts that guide citizen data scientists through the entire machine learning (ML) development process, ensuring they understand each stage of the lifecycle.

Citizen data scientists stay informed at every phase of the ML lifecycle, from data import and preparation through model building, prediction, and deployment, as the solution evolves to address their business problem.

Citizen data scientists also have visibility into the artifacts generated by the AI agent so they can review or iterate them further.

Data Import
Can I import multiple files at once? What is the file type?
Senior PM, Base Pricing
Citizen data scientists can leverage the existing SageMaker Canvas UI to select and import datasets through prompts, eliminating the need for manual uploads or extended back‑and‑forth with the chat agent.

They can also import multiple datasets from various data sources directly within the same interface generated by the agent.

Data analysis and preparation
I want to have the flexibility to review and adjust any fixes instead of having the system apply them automatically without my oversight.
Senior Data Analyst, Public Sector
Once data is imported, the AI agent analyzes it to deliver quality insights and highlight potential issues, enabling citizen data scientists to resolve errors early and iterate by selecting different target columns for prediction.

Model Building
I want an explanation on the information presented and why this model is good. Would I be able to change the metric to improve the accuracy?
Solution Architect, Public Sector
After data analysis and preparation, the AI agent builds a model from the cleaned dataset, letting citizen data scientists evaluate performance in real time without leaving the chat interface.

Citizen data scientists can also fine-tune the model on the fly without extended back-and-forth with the AI agent.

Model Prediction
I would like a quick way to test the prediction on the spot.
Senior Solution Architect, ML Tooling
After evaluating and fine‑tuning the model, citizen data scientists can generate predictions directly through the SageMaker Canvas UI within the chat, eliminating the need for manual queries or extended back‑and‑forth with the agent.

Model Deployment
I’d like to pick the right instance to deploy so I don’t end up overpaying.
Senior Solution Architect, ML Tooling
Once predictions are generated, citizen data scientists can deploy the model using the UI rendered by the agent, eliminating the need for manual queries and reducing time‑consuming, cumbersome steps.

This streamlined, end‑to‑end ML workflow provides visibility into each phase of the lifecycle, empowering citizen data scientists to accelerate decision‑making and deliver timely, high‑quality insights that solve business problems.

RESULTS AND IMPACT
The new data science assistant has been a game-changer for our analysts. For the first time, they can use natural language to leverage ML on their own, gain valuable insights into customer requirements, and making data-driven decisions without waiting on anyone. This has accelerated our time to market and enabled us to serve our customers faster than ever before.
Chief Data Officer
Financial media and technology company



