Enhancing Educational Communications: How RAG Systems Transform University-Applicant Interactions

In recent years, generative artificial intelligence has fundamentally changed how users interact with information systems, opening new possibilities in education. Higher education institutions face challenges communicating effectively with many applicants, especially during admission campaigns. Traditional counseling methods via telephone, email, or in-person meetings don't provide prompt responses and often require significant human resources.

AI-Powered Solutions for Higher Education

Developing and implementing a Q&A system based on modern language models enables round-the-clock information support, scalable service capacity, and significantly improved quality of information during admission campaigns. This approach addresses the growing need for automated solutions that can handle multiple queries simultaneously while maintaining accuracy and consistency [1].

Amazon Bedrock stands out among many platforms for developing AI applications by providing unified access to generative AI models from leading developers such as AI21 Labs, Amazon, Anthropic, Cohere, Meta, Mistral AI, and others. Key advantages include:

• Flexibility in model selection

• High scalability

• Cost-effectiveness

• Enhanced data security

• Convenient model version management

• Full integration with other AWS services

RAG Architecture and Technical Implementation

For developing an effective Q&A application for applicants, the Retrieval-Augmented Generation (RAG) architecture provides an optimal solution. This approach combines a retrieval system that finds relevant information from the university's knowledge base with a generative language model that forms natural and meaningful responses based on the retrieved data.

The system implementation consists of several key stages:

1. Information Base Preparation: Collecting documents related to admission rules, educational programs, and campus life; structuring information; and segmenting documents into logical fragments.

2. Document Vectorization: Converting text fragments into vector representations using embedding models available on the Amazon Bedrock platform.

3. Vector Index Creation and RAG Logic Development: Building a searchable vector database in Amazon OpenSearch and developing AWS Lambda functions that analyze user queries, search for relevant document fragments, form prompts for the generative model, and process responses.

Let's take a closer look at this process.

For effective information retrieval in private data, preliminary document processing is applied (Fig. 1). At this stage, documents are divided into fragments, which in turn are converted into vectors and recorded in a database, maintaining a connection with the original document. Such vectors help identify semantically similar texts.

Fig. 1. Preliminary data processing

The system's functioning mechanism is based on the principle of vector comparison (Fig. 2). When a user makes a request, the system converts it into a vector and searches for content-similar fragments by comparing vectors. Finally, the user's request is supplemented with the found context and passed to the model, which generates a response.

Fig. 2. Q&A system architecture

Performance Results and Future Developments

The developed system prototype was tested using information materials for applicants to Yuriy Fedkovych Chernivtsi National University. Research showed high accuracy of the system's responses, with an average response generation time of less than 3 seconds. The system can process numerous simultaneous requests without significant performance degradation.

Benefits for admission campaigns include:

• Reduced staff workload

• 24/7 information availability

• Consistency in provided information

• Capacity to handle hundreds of simultaneous queries

• Collection of analytical data on common applicant questions

Promising directions for further research include integration with analytics systems to identify common question patterns, enhanced response personalization mechanisms, expanded language support for international applicants, and adaptation to other educational processes.

Conclusion

Implementing RAG-based systems in higher education communication represents a significant step in the digital transformation of the educational sector. Such systems optimize workflows and improve service quality by ensuring information accessibility and accuracy. Cloud platforms like Amazon Bedrock significantly simplifies the development and implementation of such solutions, making them accessible to educational institutions of various sizes.

Further development and enhancement of these systems, particularly in personalization and multilingual support, opens new possibilities for improving the educational experience and increasing communication efficiency between universities and their potential students.

References

1. Chen, Z., Zou, D., Xie, H., Lou, H., & Pang, Z. (2024). Facilitating university admission using a chatbot based on large language models with retrieval-augmented generation. Educational Technology & Society, 27(4), 454-470 https://www.jstor.org/stable/48791566

2. Amazon Web Services, Inc. (2023). Amazon Bedrock - the easiest way to build and scale generative AI applications [Online]. Available: https://aws.amazon.com/bedrock/