cloud pipeline · BI · AI-integrated workflow
Messy operator data in. A clean warehouse, traced row by row, out.
One working slice of the stack you described: ingest from three sources, normalize and type, enrich free-text with constrained LLM structured output, and roll up into a BI view. It runs live on this page. Every value shows which stage produced it and why.
ingestnormalize / typeLLM enrichBI rollupPostgres-shaped warehouseidempotent upsertwatermark / incrementalper-row provenance
STAGE 01
Ingest
Three sources land on one natural key. Re-sends collapse.
·
STAGE 02
Normalize / type
Currency strings to integer cents. Mixed timezones to UTC.
·
STAGE 03
LLM enrich
Free-text note to a constrained category, or abstain.
·
STAGE 04
BI rollup
Null-safe aggregation by location and by category.
·