Subject-aware adaptive prompt routing framework for LLM tutoring¶
Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 causal),q2· 0 of 1 report an effect size · 2 claims rest on one study
Description¶
The article formalizes adaptive prompt selection as a contextual bandit routing problem: the router observes a pedagogical input fusing a topic embedding with a learnable subject embedding, and selects a tutoring prompt from a pool of 20 pedagogical prompts. Feedback comes from an LLM evaluator that decomposes instructional efficacy into 14 observable criteria grounded in educational theory, weighted by correlation with human labels. The authors state they "formulate adaptive tutoring as aprompt routingproblem, using topic and subject information as input."
Design Implications¶
Context¶
Requirements¶
- A pool of expert-verified pedagogical prompts
- An LLM evaluator score calibrated against human-labeled learning outcomes
- A simulation environment for safe pre-training before live deployment
Constraints¶
- Relies on a discrete prompt pool, preventing novel or hybrid instruction strategies
- Real-world data sparsity slows online convergence and limits statistical power
- Router develops bias toward strategies that perform well across the most common subjects
Target Learners¶
- Dutch high-school students
Learning Goals¶
- Subject-specific tutoring across 13 high-school subjects
- Transition to formal practice exercises
Claims¶
- Prompt Router Beats Static Baselines Simulation [+M]
- Router Subject Strategy Alignment Simulation [+M]
Related Designs¶
Examples¶
-
Key Sources¶
- Po-Chin Chang, Nicholas Hogan, Aske Plaat, Michiel T. van der Meer. (2026). Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring. https://arxiv.org/abs/2606.20138
Page metadata
| Field | Value |
|---|---|
type |
design |
id |
subject-aware-prompt-routing-framework |
title |
Subject-aware adaptive prompt routing framework for LLM tutoring |
description |
The article formalizes adaptive prompt selection as a contextual bandit routing problem: the router observes a pedagogical input fusing a topic embedding with a learnable subject embedding, and selects a tutoring prom... |
status |
review |
generated |
by process:settle-candidates on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — po-chin-chang-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.