Build your own AI film club concierge — on Databricks Free Edition, from scratch, in about an hour.
This is the complete, hands-on workshop notebook and dataset used to teach a real AI knowledge assistant build: document parsing, structured extraction, embeddings, retrieval, and a guardrailed AI agent — no prior AI or data engineering background required.
- Turning raw PDFs into structured data with Databricks AI Functions (
ai_parse_document,ai_extract,ai_classify) - What embeddings are and how semantic search actually works (vs. keyword matching)
- Building a retrieval step from scratch with cosine similarity
- Designing a reliable AI agent with a real guardrail — using structured output instead of hoping the model follows prose instructions
- Why chunking matters for real-world documents (and why we skip it here)
- Batching AI pipeline calls for real throughput — including a real bug we hit and fixed (a single unbatched query that hung for 27+ minutes, fixed by batching down to 15 seconds)
Everything in this notebook is the real technique used in production RAG/agent systems, just applied at a scale sized for a one-hour teaching session.
Lights, Camera, Agent!.ipynb— the full workshop notebook, ready to import into Databricksmovies/— the curated 50-movie dataset (one-page PDF per movie: synopsis + critic's take)
- Sign up for Databricks Free Edition — free, no credit card required
- In your workspace, go to Workspace → Import and select
Lights, Camera, Agent!.ipynb - Create a Unity Catalog Volume (the notebook walks you through this) and upload the PDFs from
movies/ - Run the cells top to bottom — every step includes a plain-English explanation of what's happening and why
Built by Sri (GitHub) — AI and data engineer working at the intersection of healthcare data and GenAI, sharing this to learn by teaching. Databricks certified via the Generative AI Engineer Associate track.
MIT — use it, adapt it, teach your own version of it.