Bad data always gets caught. The question is by whom.

DataKitchen builds open-source software for data quality testing and data observability. DataOps TestGen profiles your tables and generates the tests automatically; DataOps Observability watches freshness, volume, and schema across the estate. Both are Apache 2.0, run behind your firewall, and are used by data teams to catch bad data before it reaches a dashboard or a model.

DataOps TestGen is an open-source data quality tool that profiles every table and column, analyzes 55 column characteristics, and derives roughly 120 validation tests from what it finds. Tests run as SQL inside your own warehouse. It works with Snowflake, Databricks, Azure Synapse, Azure SQL, SQL Server, BigQuery, Redshift, Oracle, SAP HANA, and PostgreSQL.

Yes. DataOps TestGen and DataOps Observability are Apache 2.0 licensed with no feature gates, no usage limits, and no "contact sales for pricing." You install and run them yourself. Enterprise editions add support and scale features at flat pricing, but the open-source builds are complete tools rather than trials.

Testing checks the values inside your rows against rules: formats, ranges, referential integrity, business logic. Observability watches table-level behavior over time: whether the load ran, whether row counts moved, whether the schema changed. Observability catches a pipeline that silently stopped; testing catches one that ran perfectly and produced wrong numbers.

DataOps is a set of technical practices, workflows, and cultural norms that let a data team innovate quickly while keeping error rates low. It combines Agile development, DevOps automation, and statistical process control from lean manufacturing. The practical test is whether your team can ship a change today and trust the numbers tomorrow.

Yes. Alongside the software, DataKitchen provides data engineering and DataOps services: assessments that measure your current estate, consulting and coaching that changes how your team builds and operates pipelines, and training. The emphasis is on capability your team keeps rather than a delivery engagement that ends when we leave.

Yes. DataKitchen runs a dedicated commercial pharma practice at pharma.datakitchen.io, covering the data behind launch and field analytics: specialty pharmacy dispense feeds, claims, syndicated market data, 3PL reports, and patient services records. That data arrives from vendors on their schedule and in their format, which is why it breaks differently from data you produce yourself.

You did not create most of it, and the failures are relational rather than local. A prescriber or account record is assembled from several vendor feeds, so partial files, restatements that read as growth, and territory realignments all pass a null check while making the numbers wrong. Identifiers like NPI, NDC, and DEA fail silently and quietly drop rows on the join.

Install open-source TestGen, point it at one schema, and let it profile the tables and generate a baseline test set. It runs in Docker on a laptop, so there is no procurement cycle and nothing to evaluate on sample data. Most teams have a coverage number they did not have that morning within an afternoon.