AI for Science Is a Governed Loop, Not a Single Model
AI for Science is a governed workflow that combines domain skills, context, adapted models, and evaluation to turn scientific questions into reproducible evidence.
DatabricksZachary Phillips, May Merkle-Tan, Praneeth Paikray, Peter Hawkins, Vimal Thomas Joseph, Rohan Parikh, Yen LowAnnouncement
Genie Code outperforms leading coding agents by more than twice on real data tasks using Unity Catalog, but lacks healthcare and life sciences domain specificity. Vital Skills bridges this gap with packaged Agent Skills playbooks built by field engineers who require asking before assuming thresholds and using tested tools. Skills such as cohort-builder construct reproducible patient cohorts from structured and unstructured data, while open-weight-models deploys biological models with pinned dependencies. On BiomniBench, Genie Code with Vital Skills scored 76.9 on in-scope tasks versus 71.9 without, while out-of-scope tasks held steady around 71 across three blind runs. The open source skills are customizable, installable via Genie Code, and integrate with Databricks for governed deployment, model serving, and scheduling.
Genie Code, Databricks’ coding agent, uses Unity Catalog to ground responses in enterprise data, outperforming general coding agents on data tasks. However, it lacks domain-specific knowledge for healthcare and life sciences, often assuming generic methods and clinical thresholds that do not meet industry standards. This gap limits its effectiveness for biomedical analysts who need methodology-specific guidance.
Vital Skills provides packaged playbooks following the open Agent Skills standard. Each skill was built by field engineers who repeatedly solved the problem with customers. Three principles guided development: ask before assuming (pausing at key decisions like clinical thresholds), trust tested tools (using proven code instead of improvised logic), and prove effectiveness (side-by-side testing against Genie Code alone). Skills cover the HLS value chain, including cohort-builder, open-weight-models, pathway-enrichment-analysis, semantic-builder, and phi-deidentifier. They integrate with Databricks via Unity Catalog for governance, model serving, and scheduling, and are open source for customization.
Vital Skills uses an ask-before-assuming principle: at domain-specific decision points like clinical thresholds, the skill pauses and offers options for the user to choose, preventing incorrect assumptions.
On BiomniBench, Genie Code with Vital Skills scored 76.9 on in-scope tasks (up from 71.9 without) while out-of-scope scores remained stable at ~71, confirming no negative interference.
The skill set is open source, includes a self-evaluation skill (skill-eval) for benchmarking, and supports customization or creation of new skills with Genie Code’s help.
Vital Skills is a set of packaged playbooks for the open Agent Skills standard, built by Databricks field engineers to equip Genie Code with healthcare and life sciences domain knowledge. It includes skills like cohort-builder, open-weight-models, and semantic-builder, each following principles of asking before assuming and using tested tools.
On BiomniBench, Genie Code with Vital Skills scored 76.9 on relevant tasks versus 71.9 without, a clear improvement. On out-of-scope tasks, scores remained essentially unchanged (69.1 without vs 71.2 with), showing skills do not harm unrelated work.
Yes, Vital Skills is open source and hosted on GitHub. Users can install, customize existing skills, or create new ones with Genie Code's assistance. Contributions to the repository are welcome, and skills can be published to Unity Gateway for governance and sharing.
Summaries are generated from the original by a language model. Check the original before relying on it.
See the idea
AI for Science is a governed workflow that combines domain skills, context, adapted models, and evaluation to turn scientific questions into reproducible evidence.
Keep reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
One useful email. Unsubscribe anytime.