Multilingual Data Agent

This project demonstrates how to build a text-to-SQL agent capable of querying Microsoft Fabric Lakehouse and summarizing the results in natural language. It serves as a solution accelerator for scenarios where the schema or data is in a non-English language—until official support for multilingual capabilities becomes available in Fabric Data Agent.

Scenario

This demo simulates an AI assistant designed to help customer support agents, success teams, and call center representatives in an insurance company.

Agent Capabilities

• Understand user questions in Hebrew and respond accordingly.
• Interpret transaction details, insurance policies, payment statuses, and claim history.
• Automatically translate questions into SQL queries using the provided schema and run them against the Fabric Lakehouse.
• Use a query tool (FabricLakehousePlugin) to fetch results.
• Avoid asking users for additional info — rely instead on schema and predefined examples.

Table of Contents

• Multilingual Data Agent
  • Scenario
    • Agent Capabilities
  • Table of Contents
  • Features
  • Prerequisites
  • Usage
  • Contributing

Features

• 🧠 Uses Azure AI Agent Service to create and run agents in a simple, managed way
• 🧩 Extends Azure AI Agent using Semantic Kernel for tool calling and Fabric integration
• 💬 Includes a Streamlit-based sample chat UI for interacting with the agent

Prerequisites

Before running the solution, complete the following setup steps:

1. Load Data into Fabric Lakehouse
   Prepare a Lakehouse in Microsoft Fabric and upload the demo data.

2. Use Sample Datasets
   Use the datasets/ folder provided in this repo. It contains three sample tables:

   • dim_lakochot
   • dim_polisot
   • fact_tviot

3. Create a New Azure AI Agent
   Follow the official guide:
   Create and Run Agents in VS Code – Azure AI Agent Service

   Use the provided reference file setup/DataAgent.yaml as your starting point. It includes preconfigured tools and instructions aligned with this demo’s scenario and schema.

4. Clone the Repository

   git clone https://github.com/moryachok/multilingual-data-agent.git
   cd multilingual-data-agent

5. Install Python dependencies

   pip install -r requirements.txt

6. Run the Streamlit UI

 streamlit run app.py

Usage

Go to http://localhost:8501/ in your browser.

Contributing

Pull requests are welcome. For major changes, open an issue first to discuss what you’d like to change.