Skip to content
Free mini workshop · Self-paced · About 60 minutes

Intro to context engineering in Omni.

Good context does more for AI accuracy than a more expensive model. This hands-on intro shows where to add context in Omni's semantic model, and what each layer does to the AI's answers.

Hosted by Josh Bricel, founder of Spicy Data. You follow along in your own Omni instance with demo data, so every change is one you have made yourself.

Get the workshop

Free. We'll send the occasional follow-up, nothing more.

One question, before and after

"How many channel wins do we have?"

Omni Agent on the raw model: it cannot find a channel field and asks what channel wins means.
Before. The raw model has no idea what a channel win is, so the Agent guesses and asks.
After AI context, the Omni Agent recognizes channel wins, asks which time window, and for 2026 applies the business rule and returns 31.
After layer 3. Same question. Now the Agent knows what a channel win is and only asks which time window. Given 2026, it applies the business rule on its own: 31.

What you'll learn

  • Where context goes in Omni: joins, field metadata, AI context, sample queries, and the fiscal calendar.
  • What each layer does to the AI's answers. We ask the Omni Agent the same questions before and after every layer.
  • How to keep it accurate after launch with evals on a branch.

What's included

  • Six short videos, about 60 minutes total, at your own pace
  • Copy-paste YAML for every layer
  • A demo Salesforce-style dataset to load into your own Omni instance

To follow along you need an Omni instance (an active trial or customer account) and a Snowflake schema for the demo data. No Omni yet? Start a trial first.

Want to go further?

The two-day workshop does this on your own data, with your team, and ends with a measured before and after pass rate.

Start the free workshop See the two-day workshop