🚀 Databricks AppQuest Quest 5 – Intelligent Knowledge Base with RAG

[stackswift]

Hi everyone!

I'm excited to share my submission for Databricks AppQuest Quest 5.

For this quest, I extended the Lakebase starter application from Quest 4 into an intelligent Knowledge Base and note-taking application.

🔗 Deployed Application:

https://my-data-bricks-app-7474657569273252.aws.databricksapps.com/

✨ Features

• Knowledge-base note creation and management
• Markdown editing and rendering
• Note organization with tags and priorities
• Persistent data using Databricks Lakebase PostgreSQL
• RAG-based AI assistant that can use my stored notes
• Deployed application running on Databricks Apps

🧠 Prompting Approach

I first created and maintained a PLAN.md file to define the application requirements and architecture.

I then used prompts such as:

"Read PLAN.md and understand the existing Lakebase starter application and its architecture."

"Based on PLAN.md and the RAG template, create a plan to transform the existing todo application into a knowledge-base note-taking application with a RAG assistant."

For debugging and deployment, I also asked the agent to inspect errors, identify the root cause and apply the required fixes before redeploying the application.

🛠️ How I built it

I started with the Lakebase starter application from Quest 4 and used the PLAN.md workflow to define the new requirements. I then integrated the RAG functionality, updated the frontend and backend, tested the application locally, and deployed the final application using the Databricks CLI.

🎟️ Training Voucher Customization

Yes (Customized with 5 significant features: RAG AI Assistant, Markdown Editor/Preview, Tagging System, Interconnected Note Backlinks, Time Tracking).

Thanks to the Databricks and AngelHack teams for creating this hands-on AppQuest experience! 🚀

#Databricks #DatabricksAppQuest #DatabricksApps #Lakebase #RAG #AI

[ThomazNeto]

Nice work, and good call keeping a "PLAN.md" as the contract for the agent, that's the part most people skip and then wonder why the third prompt undoes the first.

Two things that might be worth a look for the next iteration, since you're already on Lakebase:

Where do the embeddings live? If they're in a separate store, Lakebase supports pgvector natively (the vector extension, 0.8.x, with ivfflat and hnsw), so a plain CREATE EXTENSION vector puts the notes and their embeddings in the same Postgres and turns the RAG lookup into a single SQL query with a join on tags and priority. Fewer moving parts in an app this size.
link1

And two things that bite demo apps after a few days: on Free Edition, apps stop automatically 24 hours after the last start or deploy, so the link you posted will need a restart from time to time. And Lakebase computes scale to zero after inactivity (default 24h) and take a few hundred milliseconds to come back, so the docs recommend connection retry logic in the app, otherwise the first query after a quiet period looks like a failure.
link2
link3

Congrats on the submission, good luck in the quest.