Paper Overview
Field: ML Authors: Quang Nguyen, Hieu Nguyen, Hien Hoang, Toan Pham, Cong Tran, Nam Vu Published: 2026-09-25 arXiv: 2609.31568
Summary
AI tutoring could markedly improve learning outcomes for students in developing regions such as Vietnam, yet the two obvious paths both fall short. Cloud assistants such as ChatGPT route sensitive student data to foreign servers — violating data-sovereignty laws such as Vietnam's Decree 53 — and, pre-trained on Western-centric corpora, are not organized around the national textbook curriculum, so their knowledge of local content is unsystematic and frequently hallucinated. Self-hosting an open model keeps data on-premise but hits a two-fold wall: post-training quantization (AWQ, GPTQ) tames the static weight footprint, yet the dynamic KV cache and prefill latency of long tutoring contexts still cause out-of-memory failures and slow responses on consumer GPUs, while the model keeps hallucinating on region-specific material.
The Approach
The authors propose DeepEdu-v1, an AI tutoring system for Vietnamese education built on the SCALE (Self-improving Context-Aware Learning Engine) framework, with two innovations:
1. Long-context inference engine: shifts token selection from per-sub-block to per-cluster granularity. In long-context retrieval, it requires 7.7x fewer retrieval calls than state-of-the-art selective-attention baselines and reduces prefill latency (TTFT) by ~35%, while task accuracy stays flat or improves. 2. Self-improving agent layer: continuously curates verified playbooks from past interactions instead of fine-tuning — a design intended to progressively reduce reliance on dominant language priors as trusted local knowledge accumulates.
Results
In its deployment configuration, DeepEdu achieves nearly 2x TTFT speedup compared to standard vLLM serving and improves agent accuracy on complex tasks from 70.0% to 79.5%, with the most significant gains on financial reasoning and interactive agent benchmarks.