Our homepage shows what a national ABA provider got. This page shows the platform itself: the architecture, what runs in production, and how every layer is governed. Written for the person who checks claims.
Sources land in bronze, conform in silver, curate in gold, and serve dashboards, Genie, and models. Unity Catalog governs every hop.
Deployed via Databricks Asset Bundles + Git CI/CD · Serverless compute · The client's own cloud
From ingestion through AI, here's what's in production today.
Streaming ingestion via Lakeflow Connect lands CentralReach, Salesforce, ADP, Zendesk, and Paradox in append-only bronze. Bronze is grant-restricted to data engineering, so raw PHI is never exposed downstream.
Silver conformed transformations feed gold curated tables, built as declarative materialized views and streaming tables (Lakeflow Spark Declarative Pipelines) on serverless compute. Identity resolution reconciles clients and employees across CentralReach, ADP, and Salesforce, so one entity gets one answer. Every transformation is version-controlled and auditable.
A standing framework of data-quality checks runs with every load, posting pass, warn, or fail to a DQ Command Center. Layer-reconciliation checks compare rows from raw through bronze, silver, and gold, so a discrepancy between systems gets caught before it reaches a report. Failures route through an acknowledge-and-triage workflow, not a Slack message that scrolls away.
"Billable utilization," "FTE," and "field care hours" are defined one time as Unity Catalog metric views and consumed identically by every dashboard, Genie space, and AI function. This is the layer that lets an LLM answer in business terms without guessing a join.
Unity Catalog end to end: column masking on PHI, attribute-based access control, full lineage. PII is classified and tagged for policy enforcement. Bronze is grant-restricted; masking applies at silver and gold. Every foundation model runs inside the client's own account, so a prompt never leaves their cloud.
Leadership asks questions in plain-English Genie spaces in production, grounded on the governed semantic layer with cited sources. AI SQL functions run inside pipelines: ai_classify rebuilt the payor and funding taxonomy, sharply cutting unclassified records, and ai_forecast projects billable-hour demand. Claude, GPT, and Gemini models are served in-account, ready to build on.
Infrastructure-as-code via Databricks Asset Bundles, Git-backed CI/CD, and a Platform Health dashboard tracking deployment and Git metrics. Dev and prod are enforced apart, with governed production objects kept separate from dev and from pipeline-internal plumbing. Inventoried, separated, documented.
A platform earns AI features in order. These are next, when the foundation justifies them.
Vector search over clinical documents, so questions like "which assessments are missing signatures?" get answers grounded in the documents themselves.
Agents that act, not just answer, running on a governed substrate that won't let them hallucinate.
Attribute-based access control, already live on core domains, expanded across the rest of the estate.
A free 30-minute review. You leave with a one-page plan, whether or not we work together.
Thanks. We respond within one business day.