Here is the English translation:

Inspiration

We were inspired by the difficulty managers and analysts have in obtaining key metrics without constantly depending on a data team. We wanted to create an AI agent that transforms natural language questions directly into SQL queries and executive analyses in real time.

What it does

OASIS is a business AI assistant that:

Receives everyday questions (e.g., "What sold the most?").
Returns the data in a table along with an executive summary tailored to the user's role.
Super-agent that branches out into 4 other agents responsible for handling different assigned tasks.
Each assignment point/part of the software is given its own solution, avoiding hallucinations.

How we built it

Java Spring Boot: Main backend REST API, packaged in Docker.
Python Microservices: Responsible for interaction with the Gemini API and query analysis.
Google Gemini API: AI model for SQL synthesis and business analysis generation.
MySQL: Relational database for storage and query execution.

Challenges we ran into

Networking between containers: Configuring internal communication between the Java Docker container and the Python services on the local host.
Character encoding: Fixing UTF-8 parsing for HTTP requests with special characters (accents and question marks) in PowerShell and Spring Boot.

Accomplishments that we're proud of

Achieving a solid hybrid architecture between Java, Python, and the Gemini API.
Containerizing the application and leaving the pipeline functional end-to-end during the hackathon.

What we learned

Docker network optimization.
Strict handling of text encoding ($UTF-8$) across multiple layers of microservices.
Prompt engineering for reliable structural SQL generation.