Meta — Daiquery/Bento notebooks
Creating unified SQL and Python notebooks to help data scientists and engineers work more efficiently with complex queries and data analysis.
I spearheaded the creation of Daiquery notebooks, a SQL cell-based system designed to help data scientists and engineers better connect, store, and understand complex queries. This work evolved into unifying SQL and Python notebooks across Meta.
Team dynamics
Worked closely with engineers, PMs, and other designers. Presented and connected across the org to get concepts moving, and partnered to maintain consistency between product spaces.
Initial creation
Work began with discovering everyday problems across the org — connecting for feedback, interviewing users, and understanding similar products in the field. Cells became a major focus, with new types added and visual treatments evolving to accommodate them.
Discovery of common feature sets
As the product expanded, we discovered commonalities with Bento notebooks (Python-based). Daiquery had file cells while Bento had file panels, which shifted discussions about where functionality should live and how cells should be structured.
Expansion into additional concepts
Cross-functional discussions explored areas missing from both products: expandable plug-in/app store, education, sharing, and more. These explorations evolved into prototypes and narratives arguing for a universal shared notebook direction.
Company changes and focus shifts
Org shifts reduced scope; broader education/documentation goals were deferred. Focus returned to the technical audience while continuing to support users. Users also used notebooks for notes, which informed upgrading text cells toward a more doc-like cell type.
AI integration
AI entered almost every discussion — surfacing join recommendations or related data objects. Explored whether users preferred in-situ recommendations vs. a helpful bot, and whether they wanted to compare results or have code produce results and move on. This included exploring concepts for AI-driven autocode completion, intelligent chart creation, and augmented data analysis workflows.
Outcome
The notebook merging worked and was used by 10k+ monthly active users at Meta. Of the concepts laid out in the original story, about 40% were realized alongside work churn. Personally, I gained exposure to AI workflow concepts that are applicable across a range of spaces — invaluable experience that informed Meta AI image creation flows and established patterns for unified notebook experiences across SQL and Python workflows.