OlehAssist

April 30, 2026 · View on GitHub

An AI-powered conversational agent designed to help new immigrants (Olim) navigate Israeli bureaucracy.

Live Demo: https://olehassist.streamlit.app

🎯 Overview

OlehAssist leverages Google Cloud Platform's Vertex AI and generative AI models to provide immediate, multilingual guidance for new immigrants to Israel. The agent helps with everything from opening bank accounts to understanding government documents and scheduling essential appointments.

✨ Key Features

  • Multilingual Support: Communicates in English, Hebrew, Russian, French, Spanish, and German
  • Retrieval-Augmented Generation (RAG): Grounds responses in official Ministry of Aliyah documentation
  • Document Analysis: Uses multimodal AI to explain government forms, bills, and letters
  • Text-to-SQL: Queries BigQuery database to find local Ministry of Aliyah branch contact information
  • Smart Routing: Prompt engineering with function calling to intelligently route users to appropriate resources
  • LLM-as-a-Judge Evaluation: Synthetic data generation to measure performance across multiple languages

🛠️ Technologies Used

  • Google Cloud Platform
    • Vertex AI (Gemini models)
    • BigQuery (structured data queries)
    • Cloud Discovery Engine (RAG implementation)
  • Python (Streamlit for deployment & Google Colab for development)

📋 Three Main Pathways

1. General Information (RAG)

Answers questions about rights, benefits, procedures using official Ministry of Aliyah documentation.

2. Document Understanding (Multimodal AI)

Upload confusing documents (bills, forms, letters) and receive structured explanations with action steps.

3. First Steps & Appointments (Text-to-SQL)

Sequential guidance through critical onboarding tasks:

  • Getting an Israeli phone plan
  • Opening a bank account
  • Scheduling Ministry of Aliyah appointment

📓 Implementation

The main implementation is in OlehAssist Code.ipynb which includes:

  • RAG system using Google Cloud Discovery Engine
  • Text-to-SQL generation for BigQuery queries
  • Multimodal document processing
  • Prompt engineering and system instructions
  • Function calling integration
  • LLM-as-a-Judge evaluation framework

🙏 Acknowledgments

Built as part of the Generative AI course at Reichman University.

📝 License

This project was created for educational purposes.