// the find
george0st/qgate-model
ML/AI meta-model, used in MLRun/Iguazio/Nuclio, see qgate-sln-<MLRun | solution>
A vendor-neutral test fixture for MLOps platforms — specifically MLRun/Iguazio — that ships a pre-built schema (feature sets, feature vectors, pipelines) plus synthetic datasets in CSV/Parquet at multiple sizes. The idea is that you run the same test suite against different ML platforms to compare how they handle feature stores, not that you build anything production-ready with it. Niche audience: MLOps engineers doing platform evaluations or regression testing between versions.
The data generation is genuinely useful — pre-bundled sizes from 100 to 100k rows covering realistic domain entities (party, transaction, account) saves real setup time. The schema is defined in plain JSON, so it's readable and diff-able without any tooling. Multiple storage targets (Redis, MySQL, Postgres, Kafka, Parquet) are explicitly tracked with completion status, which is honest. Active commit cadence with weekly activity badges that actually link to real data.
The README buries the lead badly — it takes several reads to understand this is a *test harness*, not an ML framework. The 'meta-model' framing sounds like it does something more architectural than it does. Documentation outside the README is sparse; docs/structure.md and docs/rules.md exist but aren't surfaced with any context. The generator code (`generator/synthetic_data.py` etc.) has no public API documentation, so extending it to your own schema is guesswork. 411 stars feels inflated for something this specialized — likely from MLRun ecosystem users rather than general MLOps adoption.