+
+
+
+# AI Dev Kit
+
+Give your AI coding assistant superpowers on Databricks.
+{.tagline}
+
+[Start the Quickstart :material-arrow-right:](quickstart.md){.cta-primary}
+
+
+
+
+
27Skills
+
50+MCP Tools
+
~60 minQuickstart
+
+
+
The AI Dev Kit gives AI coding assistants the knowledge and tools to build on Databricks.
Skills teach patterns. MCP Tools execute them. You talk, it builds.
+
+| Step | What you build | Time |
+|:----:|---------------|-----:|
+| **1** | Install the AI Dev Kit | 2 min |
+| **2** | Explore and profile your data | 5 min |
+| **3** | Generate a sample dataset | 5 min |
+| **4** | Build a medallion data pipeline (SDP) | 15 min |
+| **5** | Create an AI/BI dashboard | 10 min |
+| **6** | Deploy a Knowledge Assistant (RAG agent) | 15 min |
+| **7** | Build a full-stack Databricks App | 10 min |
+
+
Prerequisites: uv, Databricks CLI (authenticated), and an AI assistant (Claude Code, Cursor, or Gemini CLI).
+
+[Start the Quickstart :material-arrow-right:](quickstart.md){.cta-primary}
+
+
+
+# Quickstart
+
+##
1 Install the AI Dev Kit
2 min
+
+You need **uv**, the **Databricks CLI** (authenticated), and an AI coding assistant.
+
+=== "Mac / Linux"
+
+ ```bash
+ bash <(curl -sL https://raw.githubusercontent.com/databricks-solutions/ai-dev-kit/main/install.sh)
+ ```
+
+=== "Windows"
+
+ ```powershell
+ irm https://raw.githubusercontent.com/databricks-solutions/ai-dev-kit/main/install.ps1 | iex
+ ```
+
+Follow the interactive prompts, then open your AI assistant from the same directory.
+
+!!! success "What you'll see"
+ Your project now has `.claude/skills/` (27 Databricks skills) and `.claude/mcp.json` (MCP server config). Your AI assistant can talk to Databricks.
+
+
+
+
+
+---
+
+##
2 Explore your data
5 min
+
+Discover what's in your workspace. Open your AI assistant and paste:
+
+!!! example "Prompt"
+ ```
+ What catalogs and schemas are available in my Databricks workspace?
+ Show me the tables in each schema with their column names and row counts.
+ ```
+
+Now pick a table and profile it:
+
+!!! example "Prompt"
+ ```
+ Profile the table `main.default.my_table`. Show total rows, null counts
+ per column, value distributions for categoricals (top 10), and
+ min/max/mean for numerics. Include 5 sample rows.
+ ```
+
+!!! tip
+ Replace `main.default.my_table` with a real table. If your workspace is empty, the next step generates sample data.
+
+
+
+
+
+---
+
+##
3 Generate sample data
5 min
+
+If you need data for the rest of this quickstart:
+
+!!! example "Prompt"
+ ```
+ Generate a realistic e-commerce dataset in my workspace:
+ - main.quickstart.customers — 10,000 rows (name, email, signup_date, segment)
+ - main.quickstart.orders — 50,000 rows (order_id, customer_id, order_date, total_amount, status)
+ - main.quickstart.products — 500 rows (product_id, name, category, price)
+
+ Use realistic distributions, not uniform random. Make sure foreign keys are valid.
+ ```
+
+!!! success "What you'll see"
+ Three tables created in Unity Catalog with realistic distributions, immediately queryable.
+
+Try asking a question:
+
+!!! example "Prompt"
+ ```
+ What are the top 10 product categories by total revenue?
+ Break it down by month for the last 6 months.
+ ```
+
+---
+
+##
4 Build a data pipeline
15 min
+
+Create a production-ready Spark Declarative Pipeline with the medallion architecture.
+
+!!! example "Prompt"
+ ```
+ Create a new Spark Declarative Pipeline using Databricks Asset Bundles:
+
+ - Python (not SQL)
+ - Medallion architecture: bronze → silver → gold
+ - Serverless compute
+ - Target: main.quickstart schema
+
+ Bronze: ingest from orders and customers tables
+ Silver: clean nulls, join orders with customers, add order_year/month columns
+ Gold: materialized views for monthly_revenue (by month + segment)
+ and customer_lifetime_value
+
+ Initialize with `databricks pipelines init`, then deploy and run it.
+ ```
+
+!!! success "What you'll see"
+ The assistant scaffolds a DAB project with bronze/silver/gold Python files, deploys it, triggers a run, and shows pipeline status as each table processes.
+
+!!! info "How skills help"
+ The assistant loaded `databricks-spark-declarative-pipelines` and `databricks-bundles` skills, which taught it correct SDP patterns, serverless defaults, and Asset Bundle structure. Without these skills, it would guess — and often get it wrong.
+
+
+
+
+
+---
+
+##
5 Create a dashboard
10 min
+
+Build an AI/BI dashboard from the gold tables your pipeline just created.
+
+!!! example "Prompt"
+ ```
+ Create an AI/BI dashboard called "Quickstart: Sales Overview" using main.quickstart:
+
+ 1. Counter: total revenue
+ 2. Counter: total orders
+ 3. Line chart: monthly revenue trend (last 12 months)
+ 4. Bar chart: revenue by customer segment
+ 5. Table: top 20 customers by lifetime value
+ 6. Date range filter on all charts
+
+ Test all SQL queries before deploying.
+ ```
+
+!!! success "What you'll see"
+ The assistant follows the mandatory validation workflow: get schemas → write SQL → **test every query** → build dashboard JSON → deploy. Returns a URL to the live dashboard.
+
+Iterate by asking:
+
+!!! example "Prompt"
+ ```
+ Update the dashboard: change the monthly chart to a stacked area by segment,
+ and add a second page "Customers" with a scatter plot of order frequency
+ vs average order value.
+ ```
+
+!!! info "Why validation matters"
+ The `databricks-aibi-dashboards` skill enforces SQL testing before deployment. Without it, widgets show "Invalid widget definition" errors. Skills encode hard-won best practices.
+
+
+
+
+
+---
+
+##
6 Deploy an AI agent
15 min
+
+Create a Knowledge Assistant — a RAG-based agent that answers questions from documents.
+
+!!! example "Prompt"
+ ```
+ Create a Knowledge Assistant called "Quickstart FAQ Bot":
+
+ 1. Generate 20 sample FAQ documents (pricing, features, returns, shipping, support)
+ 2. Upload to UC volume main.quickstart.volumes.faq_docs
+ 3. Create a Vector Search endpoint and index for the documents
+ 4. Create the Knowledge Assistant using Foundation Model APIs
+ 5. Deploy to a serving endpoint
+ 6. System prompt: "Answer questions based only on the FAQ documents. If unsure, say so."
+
+ Test it with: "What is your return policy?" and "How much does enterprise cost?"
+ ```
+
+!!! success "What you'll see"
+ The assistant creates the knowledge base, vector index, and agent, deploys it, then runs test queries showing responses with source attribution.
+
+
+
+
+
+**Alternative** — for SQL-based data Q&A, try a Genie Space instead:
+
+!!! example "Prompt"
+ ```
+ Create a Genie Space called "Sales Genie" that lets users ask natural
+ language questions about the quickstart tables (orders, customers, products).
+ Add sample questions and curation instructions.
+ ```
+
+---
+
+##
7 Build a full-stack app
10 min
+
+Bring everything together in a Databricks App.
+
+!!! example "Prompt"
+ ```
+ Create a Databricks App called "quickstart-explorer" with FastAPI + React (APX pattern):
+
+ - Page 1 "Explorer": catalog/schema browser + SQL query editor with results table
+ - Page 2 "Dashboard": line chart of monthly_revenue from the gold table,
+ filterable by segment, auto-refreshes every 30s
+ - Page 3 "Chat": chat interface connected to the FAQ Bot serving endpoint,
+ with streaming responses and source document cards
+
+ Set up app.yaml with SQL warehouse and serving endpoint resources,
+ then deploy to Databricks Apps.
+ ```
+
+!!! success "What you'll see"
+ The assistant scaffolds a complete project (FastAPI backend + React frontend), configures app.yaml with resource permissions, deploys it, and returns the live app URL.
+
+
+
+
+
+---
+
+
+
+## You're done.
+
+**Data exploration** → **Pipeline** → **Dashboard** → **AI Agent** → **Full-stack App** — all through conversation.
+
+The AI Dev Kit has [27 skills](reference/skills.md) and [50+ MCP tools](reference/mcp-tools.md). Just ask your assistant to build something — skills activate automatically.
+
+!!! example "Ideas to try next"
+ ```
+ Create a scheduled Databricks job that runs my pipeline every hour
+ and sends a Slack notification on failure.
+ ```
+
+ ```
+ Set up MLflow evaluation for my FAQ Bot. Create 10 test questions and
+ measure correctness, retrieval relevance, and faithfulness.
+ ```
+
+ ```
+ Add a Lakebase PostgreSQL database to my app for storing user preferences
+ and query history.
+ ```
+
+
+
+