🧙‍♀️✨🌿 Merlin & Co. Apothecaries

Welcome! Merlin & Co. Apothecaries is a wizard-themed dbt project for the Level Up with AI workshop. It models a 15-shop potion retailer across sales, customers and guilds, and brew and shop operations.

The project runs on Snowflake with dbt v2 Stable. In this workshop you'll learn to prompt dbt Wizard to build mart models, review what it produces, and then capture your team's standards in agent files (an AGENTS.md and a reusable skill) so Wizard follows them on its own. You'll finish by rebuilding the same request in a fresh chat and seeing those standards pay off.

The loop this workshop teaches: prompt -> review -> capture -> accelerate.

Exercise outcome

Participants create:

1. models/marts/dim_shops.sql — one row per shop, built from stg_alembic_ops__shops.
2. models/marts/fct_brews.sql — one row per brew batch, built from stg_alembic_ops__brew_events.
3. Root AGENTS.md — concise always-on project instructions.
4. .agents/skills/create-mart-model/SKILL.md — the team's reusable mart workflow.

The starter project includes comment-only placeholders for both agent files. They make the intended locations visible without supplying modeling guidance before Exercise 3.

Getting started

1. Open the project in dbt Studio and confirm dbt Wizard is available.
2. Go to training_assets/exercises/ and open exercise1.md.
3. Work through the exercises in order:
   1. Build two marts with Wizard
   2. Review the first attempt
   3. Capture the team's standards
   4. Rebuild with agent instructions

Project-specific review criteria live in team notes. After drafting the agent files, compare them with the worked examples under training_assets/solution/.

Repository map

.
├── AGENTS.md                          # Exercise 3 placeholder for always-on context
├── SECURITY.md                        # non-negotiable data and action boundaries
├── .agents/skills/create-mart-model/  # Exercise 3 skill placeholder
├── models/
│   ├── staging/                       # source-facing cleanup views
│   ├── intermediate/                  # joins and grain changes
│   └── marts/                         # contracted dimensions, facts, and semantics
├── macros/                            # shared cleanup and currency macros
├── docs/merlinco/                     # ERD, data dictionary, and style guide
├── seeds/medium_data/                 # portable workshop fixtures
└── training_assets/
    ├── exercises/                     # four-part participant guides and team notes
    ├── solution/                      # model and agent-file examples
    └── reference/                     # approachable org-wide governance examples

Source systems

The 9 raw tables come from three fictional source systems:

Source	Domain	Raw tables
Abracadabra POS	Point-of-sale: potions, orders, line items, and payments	raw_potions, raw_orders, raw_order_items, raw_payments
Grimoire CRM	Wizards, guilds, and memberships	raw_customers, raw_guilds, raw_guild_memberships
Alembic Ops	Shops and brewing operations	raw_shops, raw_brew_events

Raw relations are pre-built in Snowflake and declared as dbt sources. The committed seeds in seeds/medium_data/ are portable facilitator fixtures (~15k orders, ~51k order items, 5k customers). See the data dictionary for per-table column notes and the deliberate data quirks participants will clean up.

ERD

See docs/merlinco/ERD.md for the full schema diagram with columns and types. Quick relationship overview of the raw sources:

erDiagram
    raw_customers ||--o{ raw_orders : places
    raw_customers ||--o{ raw_guild_memberships : holds
    raw_guilds ||--o{ raw_guild_memberships : grants
    raw_shops ||--o{ raw_orders : fulfills
    raw_orders ||--|{ raw_order_items : contains
    raw_orders ||--o{ raw_payments : paid_by
    raw_potions ||--o{ raw_order_items : sold_as
    raw_potions ||--o{ raw_brew_events : brewed_as
    raw_shops ||--o{ raw_brew_events : hosts

Loading

Downstream models (Kimball)

Built out already vs. what participants build in the exercises:

Layer	Already built	You build
staging	stg_ models for potions, orders, order_items, payments, customers, guilds, guild_memberships, shops, brew_events
intermediate	int_customers_with_current_membership, int_memberships_current, int_order_items_with_order_context, int_orders_with_payments, int_payments_with_order_context
dims	dim_wizards, dim_potions, dim_dates	dim_shops
facts	fct_orders, fct_order_items, fct_payments	fct_brews

Staging normalizes the deliberate raw quirks through shared macros (to_boolean, copper_to_gold, conform_region in macros/) plus direct casts and lightweight cleanup. Marts carry enforced contracts and tests.

Built-in storylines

The fixture data has patterns you can find as you explore:

• Growth: order volume climbs steadily across the two-year window (2024-07 → 2026-06); the latest half-year has ~70% more orders than the first
• Seasonality: Healing potions spike in winter, Love potions roughly double their share in February, and Luck peaks around the new year
• Regional spread: revenue varies by shop region, with about a 1.7× gap between the top and bottom regions
• Whales: a small group of customers drives outsized revenue, with the top 5% accounting for roughly a quarter of it

Support

This project is provided as-is for a workshop, without SLAs.