examples: Helm metadata-filter notebook path (PR #2195 follow-up, @jioffe502)

[kheiss-uwzoo]

Summary

Eng follow-up to PR #2195: add Helm / retriever-service guidance to examples/nemo_retriever_retriever_query_metadata_filter.ipynb.

vdbs.md#metadata-and-filtering stays a notebook pointer only — no sidecar / service prose on the docs page.

Owner: @jioffe502 (per @randerzander). cc @jdye64 if pairing is needed.

Reviewer note (@randerzander)

you don't need to update the notebook, fwiw. Jacob will do that and actually test that it's working w/ the service deployment 🙂

Docs does not need further commits on this branch. Jacob may replace or reset the opening scaffold commit when implementing.

Scope (@jioffe502)

• Add runnable Part B — Helm / retriever service cells (local Part A stays as-is)
• Service flow: POST /v1/ingest/sidecar → meta_dataframe_id on vdb_upload_params → job-based POST /v1/ingest/job + document upload → filtered Retriever.query
• Use RetrieverServiceClient / ServiceIngestor against a Helm deployment
• Validate on a real service deployment — sidecar keys in LanceDB metadata JSON, where filters work
• Reuse Part A sidecar CSV and query/filter patterns where possible

Related

• Follow-up from PR remove custom-metadata.md; canonical path is vdbs + notebooks #2195 review
• API reference: POST /v1/ingest/sidecar, POST /v1/ingest/job, test_service_sidecar.py

Test plan

• Run Part A cells locally (unchanged)
• Run Part B cells against a Helm deployment with auth configured
• Confirm sidecar keys (meta_a, meta_b) appear in LanceDB metadata JSON and where filters work

[kheiss-uwzoo]

Agreed — this is not ready for review yet.

It was opened as a tracking scaffold for the #2195 follow-up: Part B outlines the Helm / service flow (POST /v1/ingest/sidecar, job-based ingest, meta_dataframe_id) but does not yet include runnable cells. @jioffe502 and @jdye64 still need to add the executable RetrieverServiceClient / ServiceIngestor path.

The large diff is mostly notebook JSON re-serialization; the substantive change is the Part A / Part B structure and the Part B outline.

Marking as draft until the Helm cells land. Sorry for the noise — should have opened it that way initially.

[kheiss-uwzoo]

@randerzander: you don't need to update the notebook, fwiw. Jacob will do that and actually test that it's working w/ the service deployment 🙂

[kheiss-uwzoo]

Handing off per @randerzander: docs is done on #2195 / \�dbs.md. This draft is for @jioffe502 to implement and test Part B against a service deployment — no further notebook commits expected from docs. Jacob may reset or replace the opening scaffold commit.

[jioffe502]

Adding query filter options for retriever in #2258

once that merges will provide and update this notebook example

cc: @randerzander @kheiss-uwzoo @jdye64