An AI-assisted workspace for building Genies
Genie Studio gives EchoStar teams an AI-assisted way to prepare and maintain Databricks Genie spaces, which answer questions about business data. Teams describe their goal, then review the configuration AI proposes.
Sole hands-on designer for the product workflow, prototypes, component library and implementation review.
Preparing Genies for EchoStar’s business questions
Research with EchoStar analysts surfaced a problem with business definitions. For retention reporting, churn covers both paused subscriptions and disconnections. When Genie fell back to a narrower definition in a new conversation, it could leave paused subscriptions out of the count.
Four stages for preparing a Genie
Early designs for Genie Studio started with a business question. From there, teams could choose the data, check that it was ready, and test the proposed setup before creating a Genie.
Catching errors before and after creation
Before creating a Genie, teams can check the instructions it will follow and the example queries it will use. In this prototype, a review check flags a query that would count test accounts. Teams can open the editor directly from that finding to compare the suggested fix with the original query.
Once a Genie exists, teams can revise its setup and rerun a set of test questions. Saved reports record what was checked at the time. Shared scoring rules have a separate editor because changing the criteria can move a score even when answers stay the same.
In the engineering build, reviewers had to scan the configuration, score, and release panel to see what needed attention. The design update brought readiness and pending decisions together, with the full plan available for a closer look. After creation, past evaluations got their own page, and saved reports put blocking checks up front so reviewers could open the supporting evidence.
Finding the detail behind a metric
Monitoring helps people go from a portfolio overview to a question about one Genie. Filters set the space and time period across the dashboard. Summary cards link to their detailed metrics, and search jumps directly to a chart or table.
Monitoring walkthrough
This video shows how to narrow the dashboard to one Genie space, then use the question log to see which requests came back empty.
A coded library for EchoStar’s design system
EchoStar’s existing design system was built into a reusable code library, with adaptations for the Databricks controls already used in Genie Studio. Working components were added to Storybook by hand so designers and developers could inspect behavior outside the application.
Side-by-side comparisons in Review Lab show how a component behaves in the original Figma design, the code library, and the Databricks toolkit. Matching the size and state across all three gives reviewers a way to spot implementation differences before discussing design changes.
Consistent headers and cards across Genie Studio
A Genie’s name and tabs stay in the same place across setup, evaluation, and saved reports. Each page keeps room for its own actions.
In setup and evaluation, review cards place the title and finding above the actions, so reviewers have a familiar place to look for the next step. Compact cards use tighter padding, and when space is limited, the buttons stack in the same order.
From business questions to a Genie Agent
This walkthrough shows how to set up a Genie Agent to answer questions about franchise sales. After choosing data sources and reviewing the proposed setup, teams can save a draft or create the Agent in a test environment. As EchoStar’s business teams begin creating Agents for their day-to-day work, the project team will monitor how they use Genie Studio and refine it based on those observations and direct feedback.