This demo is based on the initial trino setup made here.

Once you setup trino on your local k8s cluster, you can continue with this demo.

This is highly experimental, and VannaAI does not explicitly support trino, but I have worked around it with their run_sql functionality.

• trino see comment above
• python3.11
• uv
• make
• gemini (free tier)

Since I run on mac, I had to pin:

kaleido==0.2.1 to avoid issue:

When installing vanna it uses 0.2.1.post1. Hence giving error:

• error: Distribution kaleido==0.2.1.post1 @ registry+https://pypi.org/simple` can't be installed because it doesn't have a source distribution or wheel for the current platform`

Head out to google AI Studio And Create an API key: https://aistudio.google.com

Once you have this you need to make sure either globally, or within your session you have:

export GEMINI_API_KEY=...
export GEMINI_MODEL=... # example gemini-2.0-flash

The rest of the demo assumes this.

Fact table of video playback events. References dimensions: dim_videos (video_id), dim_categories (category_id), dim_creators (creator_id), dim_users (user_id).

Dimension table describing videos

Dimension table categorising video content

Dimension table containing video creator details

Dimension table describing user/viewer metadata

                  dim_categories
                        │
                        │
dim_creators─────fact_video_plays─────dim_videos
                        │
                        │
                    dim_users

uv venv

make setup-environment

make run-app ARG=--train

make run-app

• local vanna ai setup