Build a Text-to-SQL AI Agent on Databricks — 2-Hour Bootcamp

In 2 hours, you'll build and deploy a production-style text-to-SQL AI agent on Databricks — complete with SQL safety guardrails, schema-aware prompts, and live deployment. Capped at 20 seats so every participant ships working code.

This is a hands-on engineering session for data engineers, data analysts, AI engineers, and platform teams who want to go beyond demo notebooks and ship internal AI tools that respect data governance, query safety, and production realities.

What you'll build

A working AI agent that:

• Converts natural language questions into safe SQL
• Validates queries against your Databricks schema with SELECT-only guardrails
• Executes against a Databricks SQL Warehouse on top of a real datalake (Delta tables, Unity Catalog)
• Returns results with explanations the requester can audit

You leave with a forked, configured repo you can deploy at your workplace on Monday.

Why this bootcamp exists

According to the Databricks 2026 State of AI Agents, 80% of new enterprise databases are now generated by AI agents. Every data team is being asked to build one. Most teams get the LLM call right and the safety + governance layer wrong — exposing production datalakes to destructive queries, accidental PII access, or hallucinated joins.

This session fixes that. The focus is the boring, important parts:

• Prompt engineering with schema context — so the model knows your tables
• A SQL parser-based validator that allows only SELECT and rejects mutations
• Row- and column-level guardrails compatible with Databricks Unity Catalog
• Audit logging for every agent query — the foundation of data governance for AI tools
• Patterns that work whether your back end is Agent Bricks, a Databricks Job, FastAPI, or AWS Lambda

You'll leave with

• A deployed text-to-SQL agent on Databricks (your own workspace or ours)
• The open-source db-agent repo cloned and configured
• A working SQL safety validator (SELECT-only, schema-enforced)
• The full prompt-engineering pattern for schema-aware generation
• Deploy-ready patterns for Streamlit, FastAPI + Next.js, and AWS Lambda

Who it's for

• Data engineers building internal tools on top of Databricks, Delta Lake, and Unity Catalog
• Data analysts who want a self-serve SQL assistant for their team that doesn't go off the rails
• Data governance leads evaluating how AI agents can safely query production datalakes
• AI / platform engineers designing agentic workflows over enterprise data
• Technical leads evaluating text-to-SQL and Agent Bricks for their organization
• Solutions architects at consultancies and partners shipping Databricks-based AI products

If your job touches a Databricks workspace, a datalake, or anyone asking "can the AI just query this for me?" — this is the session.

Prerequisites

• Comfort with Python and SQL
• A laptop and a GitHub account
• A Databricks workspace is helpful (we'll provide a shared sandbox if you don't have one)

That's it. No prior experience with LLMs, agent frameworks, or RAG required.

Format

• Live on Microsoft Teams — interactive, video on, screen sharing welcome
• Saturday, May 16, 2026 — 12:00 to 2:00 PM EDT (Toronto time)
• Capped at 20 seats so every participant gets attention
• Recording sent to attendees within 48 hours
• Discord channel for follow-up questions and code review

About your instructor

Chandan Kumar — Founder of beCloudReady (Databricks Registered Partner) and organizer of TorontoAI, a 10K+ member community of AI builders.

Author of the open-source db-agent project — presented at the AAAI-25 workshop on AI agents — which is the foundation we'll build on during the session. Twenty-plus years in software engineering, cloud architecture, and data engineering across enterprises in Canada, the US, and the UK.

Resources we'll use

• github.com/db-agent/db-agent — the open-source repo
• quick-labs.co — sandbox Databricks environment for participants without a workspace
• toronto-ai.org — community for follow-up discussion
• becloudready.com — longer-form bootcamps if you want to go deeper after this

Why "Agent Bricks" matters here

Databricks recently introduced Agent Bricks as the native way to build, evaluate, and deploy AI agents on the Databricks platform. The patterns we cover in this bootcamp — schema-aware prompting, query safety validation, audit logging — translate directly to Agent Bricks deployments, but the principles are platform-agnostic. Whether you ship on Agent Bricks, a containerized FastAPI service, or AWS Lambda, the engineering is the same.

Build something real. Not just prompts.

There are plenty of "build an AI agent in 5 minutes" YouTube tutorials. None of them deploy something a production data team would actually trust against a real datalake. This session does.

Reserve your seat — 20 spots only.