Open Category Submission

722 – LLM Assistant for High School English Composition Correction – Thinking Process and Prototype Construction

In non-English-speaking countries or regions with limited English teaching resources, the correction of high school English compositions is mainly carried out by classroom teachers. Traditionally, teachers individually correct and grade students' responses to fixed topics, with the correction focusing mostly on language basics such as grammar and vocabulary spelling. However, in follow-up teaching activities with the assistance of innovative AI assistant, teachers can summarize the common errors and have an overall understanding of students’ compositions in a blink of an eye.
This is an intelligent composition correction and optimization assistant for high school English teaching scenarios. With profound knowledge and rich experience, it can conduct page-by-page detailed analysis, accurate scoring, in-depth correction and optimization of high school English composition topics and answers in PDF format according to users' set scores and requirements, draft revised paragraphs, and integrate them into a standardized PDF feedback document. The objectives include assisting teachers in completing composition correction work and improve work efficiency, providing targeted modification suggestions and optimized paragraphs, and enrich educational resources and offer English writing guidance without interpersonal barriers or high tuition fees.
The AI assistant consists of six chosen parts, containing agent service provider, large language model, knowledge base, plugin, prompts of restrictions and prompts of skills. Four technical highlights and implementation details are referred: Analyze compositions &Sample Composition Writing; Composition Scoring; Correction and Optimization; Overall Analysis.
We have decided to carry out further development work in the future. Complex functions described will be decomposed, and the decomposed functions will be individually partitioned into separate Agent intelligent entities. Local programs and corresponding API interfaces will be used to achieve combined invocation of single functions, thereby implementing the LLM Agent for high school composition review described in this document.

The inspiration for developing this AI essay grading tool stemmed from my firsthand experience as a high school English teacher. Year after year, I faced the arduous task of manually grading numerous essays, spending excessive time on repetitive tasks like checking grammar, spelling, and vocabulary usage. I noticed that students often had to wait days for feedback, which hindered their learning progress. Moreover, I realized that many students made similar errors, and a consistent, efficient grading system could help address these issues promptly. The hypothesis underlying this project is that AI, with its advanced natural language processing capabilities, can accurately assess English compositions based on predefined criteria, such as grammar, coherence, and content relevance. I believe this tool will succeed because it can provide instant feedback, allowing students to correct their mistakes immediately and learn more effectively. Additionally, it can reduce teachers' workload, enabling us to focus more on individualized instruction and deeper discussions.

To implement the solution, we will utilize a large language model (LLM) to analyze, score, correct, and optimize high school English compositions. The agent service provider, along with plugins, prompts of restrictions, and prompts of skills, will guide the LLM's operations. The knowledge base will offer reference for grammar, vocabulary, and writing norms. We need technical talent proficient in NLP and LLM development, computing resources for model training and deployment, and a dataset of high school English compositions for validation. To gauge market demand, we'll survey teachers in regions with scarce English teaching resources and conduct trials in local schools. The core functionalities include detailed page - by - page composition analysis, accurate scoring, in - depth correction, and optimization, along with generating revised paragraphs and standardized PDF feedback. We'll ensure a positive user experience by providing instant feedback, easy - to - understand suggestions, and a user - friendly interface. Effectiveness will be evaluated using metrics like the accuracy of error detection, the relevance of improvement suggestions, and teachers' and students' satisfaction rates after using the tool.

Functional Architecture and Technical Workflow
The project's functional architecture consists of components like the Large Language Model, Knowledge Base, Plugin, Prompts, and Agent Service Provider. The technical workflow starts with PDF reading. Then, functions such as Sample Composition Writing, Composition Scoring, Correction and Optimization, and Overall Analysis are executed. For instance, in Sample Composition Writing, relevant knowledge from the knowledge base is used to create a sample essay. After all analyses, the results are integrated into a PDF for output.
Implementation of Innovative Features
Innovative features like accurate scoring and in-depth correction are implemented using advanced grammar-checking tools, vocabulary recommendation systems, and corpus. The scoring model in Composition Scoring ensures fairness. For Correction and Optimization, the system provides multiple improvement suggestions with explanations.
Design and Development Timeline
The initial version was tested in April 2025. Future plans involve decomposing complex functions into separate Agent intelligent entities. Performance metrics include grading accuracy compared to human teachers and the correlation between agent and teacher grades.
Relationship between Functions and Technologies
The functions rely on technologies like the Large Language Model for analysis and writing, and the knowledge base for reference. Currently, the single - Agent deployment has limitations, but the future multi - Agent design aims to overcome them.

The idea of this AI - based high school English composition correction assistant is innovative in several ways. Currently, manual composition grading by teachers in regions with limited English teaching resources has issues like low efficiency, inconsistent scoring, and lack of personalized feedback.
This project uses a large language model, knowledge base, and plugins to provide an automated solution. It can quickly analyze, score, correct, and optimize compositions, which is a significant improvement over manual work. For example, in Composition Scoring, it uses a scoring model to ensure fairness, and in Correction and Optimization, it offers multiple suggestions based on corpus and writing template libraries.
These innovative elements enhance its effectiveness. Teachers can save time and get more accurate and consistent scoring. Students receive personalized feedback immediately, which helps them improve their writing skills. Overall, it bridges the gap in English teaching resources and provides a more efficient and effective learning and teaching experience.

To ensure scalability, we'll decompose complex functions into separate Agent intelligent entities. This modular approach allows for easier expansion as each Agent can be independently scaled based on demand. We'll use local programs and API interfaces for combined function invocation, reducing the load on a single system. Potential bottlenecks, like high text volumes in PDFs causing incomplete function implementation in multi - Agent mode, will be addressed by optimizing data processing algorithms and ensuring efficient communication between Agents.
Regarding environmental sustainability, we'll aim to use energy - efficient computing resources and optimize the model's performance to reduce power consumption. To foster long - term user engagement, the assistant will continuously update its knowledge base to provide the latest language knowledge and writing styles. We'll also improve the user - friendliness of the interface. To adapt to evolving user needs, we'll collect user feedback regularly and make iterative improvements to the functions, such as adding new types of writing analysis based on emerging teaching requirements.

This solution addresses social issues by alleviating the burden of English composition grading in regions with limited teaching resources. For teachers, it boosts efficiency, allowing them to focus on more personalized instruction. For students, it provides timely, targeted feedback, enhancing their learning experience and writing skills, which is crucial for academic progress.
In terms of equity, the tool offers consistent grading and equal access to high - quality writing guidance regardless of students' backgrounds or locations. This aligns with the broader social goal of inclusion in education.
To measure social impact, we'll use metrics like the improvement in students' writing scores over time, teachers' satisfaction with the tool's efficiency, and the reduction in the gap between students' writing abilities. We'll ensure responsiveness to community needs by regularly collecting feedback from teachers and students. We'll analyze this feedback to identify areas for improvement, such as adding more diverse writing topics or adjusting the scoring criteria, and then make iterative changes to the tool.

Personal Information Collection Statement (PICS):
1. The personal data collected in this form will be used for activity-organizing, record keeping and reporting only. The collected personal data will be purged within 6 years after the event.
2. Please note that it is obligatory to provide the personal data required.
3. Your personal data collected will be kept by the LTTC and will not be transferred to outside parties.
4. You have the right to request access to and correction of information held by us about you. If you wish to access or correct your personal data, please contact our staff at lttc@eduhk.hk.
5. The University’s Privacy Policy Statement can be access at https://www.eduhk.hk/en/privacy-policy.

• I have read and agree to the competition rules and privacy policy.