Researchers had data but not answers, and custodians were afraid to give them direct access. Questions is the layer between raw tables and usable answers: custodians publish a safe, parameterized question once, and researchers get meaningful answers without wading through raw data or seeing sensitive patient information.
A single collection can hold hundreds of tables, and each table can hold millions of variant records. One table I worked with had about 49 million rows. Researchers had the data but not the answers, and custodians couldn't safely open it up.
Hundreds of tables, each with millions of variants, and no way to ask a clinical question of them.
Is this variant rare? Benign or pathogenic? Linked to disease? Raw rows can't say.
Direct table access risks exposing identifiable patient information.
Researchers need clinically meaningful questions, not raw database rows.
We want to support research, but we can't risk exposing sensitive patient data.
What was missing: an abstraction layer between raw tables and usable answers. Raw data isn't research insight, and custodians need safety and control.
My work was everything that turned Questions into a product people could trust: the two user journeys, the custodian safety model, the researcher experience, and the post-launch roadmap. I mapped the end-to-end research workflow across lab technicians, bioinformaticians, principal investigators and clinical geneticists to find where Questions fit in their day.
Custodians do the hard work once. Researchers get meaningful answers without wading through raw tables or seeing sensitive patient information.
The variable system is what makes a question reusable across both journeys. I prototyped the end-to-end flow, from creating a question in Publisher to querying it in Explorer, before it went to engineering.
The feature validated DNAstack's platform as a globally viable research tool.
of sessions came from outside North America
users returned within the same month
of users stayed longer than 30 minutes
of users found the published questions highly relevant to clinicians
After the MVP launched, I took the lead on improving Questions. I ran feedback sessions, empathy mapping and how-might-we workshops with researchers and custodians. Two audiences came through: technical users who wanted scale, and custodians who weren't comfortable writing SQL. Each need became a stage of the roadmap, agreed in tradeoff discussions with product, engineering and the bioinformaticians.
"I would like to ask questions at scale. Querying 1000s of variants one at a time slows everything down."
I worked with engineering to shape a CLI and Python library so researchers could list questions, read their parameters and run them in bulk from their own pipelines. It came first because it extended questions custodians had already published, with no new privacy risk.
"Honestly, I wish there were templates for common research questions. I'd rather configure than code."
Not every custodian was comfortable writing SQL. I worked with the bioinformaticians to find scientific questions that hold up across many datasets, like allele frequency and carrier counts, and turned them into templates custodians configure instead of code.
"I know the data really well, but I'm not confident writing SQL joins. I worry I'll make a mistake and expose sensitive fields."
Templates cover common questions, but non-technical custodians still needed a way to write their own. An AI assistant in the query editor drafts SQL from a plain-language request. We placed it after templates because a drafted query that exposes a sensitive field is exactly what custodians feared.
