Key takeaways
- MariaDB Cloud simplifies AI development by managing structured data, geospatial records, and vector embeddings within a single, unified database.
- Real-world demos show how "semantic search" understands user intent and context, providing more accurate results than traditional keyword-based searches.
- Tools like MariaDB Serverless bridge the gap between technical data and business "why," allowing even non-developers to build intelligent applications that join deal history with unstructured notes.
As developers, we are skeptical of “AI marketing”. We want to see it run. We want to see the actual techniques and features in action. So, at MariaDB, we decided to organize an AI Demo competition inviting coders (and also non-coders!) to solve actual problems using our cloud and AI capabilities.
The results are really cool! We saw applications handling geospatial data, analyzing business logic, and even navigating “vibe coding” workflows. So, let’s explore some of these demos with links to video and source code repos.
1. Metric Moon: The “Why” Behind the Data
Traditional BI dashboards are great at showing you what happened but not why it happened. This demo builds an analytics app that queries across both structured and unstructured data. For example, you can ask “why are outdoor products underperforming?” and the application will execute a SQL query to pull the raw sales metrics and run similarity search against vector-indexed marketing reports, combining both sources into a single answer.
The app uses MariaDB Cloud Serverless to store both the structured sales data and the vector embeddings of the business documents.
Source code: https://github.com/mariadb-philip-ganchev/metric-moon
Video: Watch the Metric Moon demo on YouTube
2. Austin City Explorer: Spatial + Vector Synergy
Evaluating a neighborhood requires querying multiple distinct data types: spatial data (crash reports, tree inventories) and unstructured text (zoning PDFs). Traditionally, this means maintaining and orchestrating separate GIS databases and document stores. This demo provides a unified retrieval layer that handles both geospatial queries and semantic search in a single pass.
The app uses MariaDB vector and geometrical capabilities (like spherical geometry spatial functions) both available in MariaDB Cloud. The database holds 2.2 million construction permits, 62,000 tree records, and hundreds of vectorized city planning PDFs.
Source code: https://github.com/mariadb-BenGrieser/mariadb-ai-demo-austin-real-estate
Video: Watch the Austin City Explorer demo on YouTube
3. The AI Sales Co-Pilot: “Vibe Coding” a CRM Tool
Sales datasets typically live in multiple systems. This makes it difficult to query qualitative information programmatically.
This app was built by a sales leader, not a full-time developer! The “Sales Co-Pilot” mixes and joins deal history with unstructured notes. It can look at a current deal and find similar past deals based on “semantic meaning” rather than just keywords. This is useful for identifying risks based on historical patterns.
Source code: https://github.com/buddyosher/MariaDB_XANO_Bolt
Video: Watch the Sales Co-Pilot demo on YouTube
4. The “Extreme” Outdoor Shop: Semantic Search for Context
Standard lexical search has low recall when the user’s vocabulary differs from the indexed text. For example, a standard query for “trip to the Alps” will fail to return relevant items described with conceptually related tokens like “skiing” or “cold weather” without maintaining extensive, hardcoded synonym dictionaries.
This demo app understands context. When a user asks for gear for a “trip to the bottom of the Mariana Trench,” the search engine recommends an oxygen suit and flares even though the user never mentioned “diving.” The app also showcases how to handle “prompt injection” to avoid malicious actors.
Source code: https://github.com/kpac-mdb/MariaDB-Vector-Search-Demo
Video: Watch the Extreme Outdoor Shop demo on YouTube
Conclusion
These demos prove that building AI-powered applications doesn’t require a complex architecture of five different niche databases. Even if you are doing spatial analysis on millions of records, building a business intelligence tool, or creating a sales assistant.
Check these and other videos on YouTube and let us know what you think!
