Development of a Hyper-scale AI-Based Service for Early Identification and Support of Slow Learners

Overview

As the number of borderline intelligence individuals (slow learners) increases, social issues related to childcare in homes and educational institutions are intensifying. Establishing a proactive prevention system to detect these issues early is crucial. The service utilizes cutting-edge AI technologies such as RAG (Retrieval Augmented Generation) and LLM (Large Language Model). This service assists parents in recognizing their children's developmental and learning challenges early on and enables timely intervention. It is expected to minimize learning gaps for slow learners and contribute to ensuring more children receive appropriate support.

Project description

The goal is to build an optimized AI system through data accumulation and processing, AI model training, and validation to develop a service supporting the early detection of slow learners. First, behavioral data of slow learners will be collected and processed to build a RAG-based question- answering dataset. Subsequently, pre-training and fine-tuning will be performed to enable effective LLM learning, followed by iterative validation and improvement processes to enhance the model's response accuracy.

The primary service processing method operates by having RAG retrieve relevant information based on questions entered by parents, with an LLM then providing customized responses based on this information. The analysis targets key cognitive domains such as language, memory, perception, concentration, and processing speed. It supports real-time counseling by linking a FastAPI-based server with the frontend. This enables the early identification of children requiring learning support and assists parents in implementing appropriate interventions.

Key technological features

• Establishing a system leveraging RAG (Retrieval Augmented Generation) and LLM (Large Language Model) to support the early detection of slow learners.
• The RAG module retrieves relevant information in response to user queries, while the LLM generates personalized responses based on the retrieved content.
• Providing analysis by major cognitive domains (language, memory, perception, attention, processing speed), and implementing real-time consultation by linking a Fast-API-based server with the frontend.

Expected outcomes

The service enables parents to more easily understand their children's learning difficulties and receive expert-level advice, thereby improving the home learning environment. For public officials and educational institutions, accumulated data can be leveraged to more effectively identify children requiring policy support, enabling the establishment of personalized educational policies. Furthermore, this system enhances accuracy through continuous data learning, increasing its scalability for future use in education and counseling fields. Based on these achievements, it is being supplied to up to 200 caregivers within the Seoul Metropolitan Government's Borderline Intelligence Program from 2024 to March 2025.

Launch year: 2024

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.1,  https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=28997&parentSeq=28998