giva-databricks 0.1.0

One-line installer for the GIVA Databricks demo (AI jewelry commerce: data + Vector Search + Genie + AI/BI dashboard + Lakebase + FastAPI app + paused background jobs).

💎 GIVA — one-line Databricks demo installer

giva packages the entire GIVA AI jewelry-commerce demo into a single pip-installable package, à la dbdemos. One call provisions the whole stack into your workspace from pre-baked real data.

%pip install giva-databricks
dbutils.library.restartPython()

import giva
giva.install('giva', catalog='my_catalog')

Locally with a CLI profile:

pip install giva-databricks
python -c "import giva; giva.install('giva', profile='AnujLathi', catalog='anuj_vm_workspace_catalog')"

What gets installed

#	Asset	Detail
1	UC data layer	schema + volume, enriched_jewelry_products (~236 LLM-enriched products) + jewelry_embeddings + ~236 product images
2	Vector Search	giva-vs endpoint + jewelry_embeddings_index (DELTA_SYNC, 1024-dim) for semantic & image search
3	Genie Space	natural-language analytics over the product catalog
4	AI/BI dashboard	catalog & sales analytics, published
5	Lakebase (Postgres)	giva-orders instance + giva db: orders, users, nudges, live metal prices, dynamic product prices — seeded with data
6	App	React + FastAPI storefront + admin analytics (Databricks App)
7	Background jobs	Metals Refresh + Nudge Emails — created PAUSED (stopped, not triggered)

⏸ The two pipelines are installed PAUSED

Per design, the Metals Refresh Pipeline and Nudge Emails Pipeline jobs are created with their schedule pause_status = PAUSED. They will not run on a trigger. The demo works out of the box because metal prices and product prices are seeded into Lakebase. When you want a live refresh, either:

• trigger them manually from the admin app (Run now), or
• unpause the schedule in the Jobs UI.

The two job notebooks read Service-Principal credentials and (for emails) Gmail OAuth from a Databricks secret scope giva. These secrets are not created by the installer — set them before unpausing if you want the jobs to run.

Options

giva.install(
    'giva',
    catalog='my_catalog',          # required on non-Free-Edition workspaces
    schema='giva',          # default
    warehouse_id=None,             # default: first running/available
    install_app=True,
    install_lakebase=True,         # provision Lakebase + seed
    install_jobs=True,             # create the two PAUSED jobs
    install_vector_search=True,
    profile='AnujLathi',           # only when running locally
)

Rebuilding the package data (maintainers)

The packaged data/, dashboards/ and Lakebase schema are exported from the live reference build. To refresh them:

python scripts/export_live_assets.py --profile AnujLathi \
    --catalog anuj_vm_workspace_catalog --schema caratlane_jewelry
python -m build           # produces dist/giva-*.whl

Installing the package

Until it's on PyPI, install the wheel straight from a GitHub Release:

pip install https://github.com/anuj1303/giva-demo/releases/download/v0.1.0/giva-0.1.0-py3-none-any.whl

or build-and-install from source:

pip install "git+https://github.com/anuj1303/giva-demo.git"

Publishing to PyPI (maintainers)

The Publish giva-databricks to PyPI GitHub Action publishes via trusted publishing (OIDC — no API tokens). One-time setup, then a one-click publish:

1. On https://pypi.org → your account → Publishing → Add a pending publisher:
   • PyPI Project Name: giva
   • Owner: anuj1303
   • Repository name: giva-demo
   • Workflow name: publish.yml
   • Environment: (leave blank)
2. In the repo: Actions → Publish giva-databricks to PyPI → Run workflow.

Note: the wheel is ~88 MB (the 11.5M-row nudge_emails seed is most of it). That's under PyPI's 100 MB per-file limit but large — the GitHub Release wheel above is the recommended distribution for most users.

Reference build

• Workspace: AWS FE VM (fe-vm-anuj-vm-workspace)
• Source catalog/schema: anuj_vm_workspace_catalog.caratlane_jewelry
• App: tanishq-jewelry-demo (UI name GIVA)