Read what the dashboards need. Build the layer that feeds them.
The stack runs Source → Airbyte → Snowflake → dbt → Sisense. This portal writes the dbt layer for you: it reads the Sisense intent, studies the raw landings, asks the few questions a consultant would ask, and generates transformation SQL it can prove correct — validated cell-for-cell against a live production layer.
Connect a source
Add a project: the Snowflake connection where the raw data lands (Airbyte connectors — Infor ERP, Salesforce, Procore, databases — all land there), plus the target layer to build toward. Credentials are encrypted; a live test verifies access before you save.
Add the dashboard intent
Drop the Sisense widget exports (*.jaql.json) on the Create Layer page. Each becomes a precise contract — fields, grain, aggregations, filters — and the tables the dashboards need appear automatically.
Build & verify
The AI asks anything it would otherwise guess (once, ever — answers persist as conventions), writes the dbt SQL blind from intent + raw source, self-heals against answer-key-free checks, then reveals vs production for a cell-for-cell GO/HOLD verdict.
Why this is trustworthy
Blind by construction
While building, the AI never sees the existing dbt definitions or the production output — only the dashboard intent and the raw source. It converges using answer-key-free validators: grain uniqueness, source control totals, intent coverage.
Proven on a live tenant
On the DC engagement the generated layer matched the hand-built production dbt table at 99.9998% cell-for-cell, with every Sisense KPI recomputing identically. Confirmed answers persist as per-project conventions, so accuracy compounds with use.