Headspace provides mental health support to millions of members worldwide. As the organization expanded, managing bespoke reporting tables across disparate tools grew increasingly complex. To establish a single source of truth, Headspace unified 13 domain schemas on the Databricks Data + AI Platform on AWS. By deploying AI-powered conversational analytics through Genie Agents, Headspace enabled self-service data exploration while maintaining centralized governance.
Streamlining data modeling for metric consensus
Headspace continuously seeks to innovate its mental health ecosystem. Over time, as business needs evolved, earlier data modeling iterations led to custom tables scattered across multiple environments. Data processing was distributed across various tools, including dbt, Prefect and custom Docker containers. Because these tools operated outside a unified governance boundary, definitions for critical business reporting, such as registrations, sign-ups and subscriptions, began to diverge across business units and engineering teams.
Leadership brought business and technical stakeholders together to establish clear ownership for core metrics and set accuracy thresholds based on operational impact. Once consensus on metric definitions was established, Headspace turned to Databricks to construct a flexible foundation built for conversational AI.
"AI will not rescue an ambiguous definition," said Mary Alfheim, VP Data & Analytics at Headspace. "It will improvise one for each person who asks, generating contested numbers faster. We aligned our people on core business metrics before writing a single line of code."
Implementing a contract-driven data model with Databricks
With business consensus established, Headspace adopted a contract-driven generative architecture on the Databricks Data + AI Platform on AWS. This standardized modeling, governance and downstream consumption. Governed data contracts written in versioned Protocol Buffers became the mechanical input for the entire warehouse.
From contract to warehouse
A custom projection engine automatically generated SQL directly from these contracts. It also generated pipelines, orchestration graphs and quality expectations to help teams scale changes without re-creating logic across tools.
Headspace replaced manual configuration with native Databricks capabilities. Declarative Automation Bundles emitted deployable infrastructure, and Lakeflow Jobs orchestrated workflows from compiled graphs. Expectations validated incoming data streams.
Benefits of a governed data model
Unity Catalog established centralized access control across 13 domain schemas and roughly 49 conformed entities. It also propagated sensitivity tags down the lineage graph to preserve HIPAA compliance.
"Architecting backward from contracts gave us the velocity to rebuild our enterprise data model in three months," said Marlissa Wong, Director of Data Engineering at Headspace. "With Declarative Automation Bundles and Unity Catalog, edits to a contract regenerate the pipeline automatically, allowing our team to focus on business value rather than infrastructure."
Enhancing organizational insights with Genie Agents
To give teams instant, reliable access to AI-driven insights, Headspace built Aladdin, a governed driver agent powered by Claude. Aladdin serves as an intelligent router, identifying domain context and directing questions to the appropriate agent. It sits directly above 11 domain-specific spaces powered by Genie Agents, so users no longer need to know which schema holds the answer.
A wide, shallow routing agent enables the Genie Agents to become narrow, deep domain experts. Aladdin passes each question to the Genie Agent best equipped to answer it. That agent translates the natural-language prompt into transparent SQL, executes queries and calculates precise metrics. Business definitions reside in a single context layer in code. This ensures that an analyst, product manager or executive asking the same question receives identical results. Today, more than 30 active users query governed data through Aladdin, up from six analysts who previously handled data requests.
The contract-driven foundation and Genie Agents delivered measurable gains across Headspace:
Data contract modifications accelerated from a full quarter to under a day: Engineers updated versioned contracts to regenerate deployment pipelines automatically.
Data modeling modernized into 13 conformed domain schemas: Standardized Kimball modeling frameworks replaced wide report tables with roughly 49 conformed entities.
Governed metadata standardized across 178 versioned contracts: Structured protobuf contracts replaced roughly 4,000 loose description strings.
Self-service analytics expanded to more than 30 active users: Business users query data safely in natural language without accessing raw staging environments.
"Aladdin answers where data lives, and Genie shows you the exact SQL," said Joseph Kroon, Data Architect at Headspace. "Anyone can ask how a metric is defined and get a clear answer because definitions live in a traceable repository rather than individual queries."
FAQ: Headspace on Databricks Genie Agents
Headspace uses a driver agent, Aladdin, to route questions to 11 domain-specific Genie Agents for governed, self-service natural-language analytics.