TmaxEduAI

Recommendation Systems, Knowledge Tracing

EdTech

Developed AI-based educational services and solutions to enhance personalized learning experiences.


Summary

Role

Led the development of AI-based projects for education, focusing on Knowledge Tracing and Recommendation Systems.

Experience

  • Developed deep learning models for AI-driven education services using sequential interaction data.
  • Designed and implemented automated ML pipelines for scheduled training and batch inference.
  • Addressed the cold-start problem in Knowledge Tracing and Recommendation Systems.
  • Designed data models and developed APIs.

Tech Stack

  • Programming Languages:  Python, SQL, Java
  • Frameworks & Tools:  PyTorch, Airflow




Projects

AI Digital Textbook

Description

Developed core features for the assessment and recommendation systems in a digital textbook. Designed a knowledge tracing formula to estimate students' knowledge states, enabling data-driven educational decisions, and developed rule-based problem recommendation algorithms based on knowledge state levels.

Contributions

  • Analyzed customer requirements, set sprint goals, and managed development progress to ensure alignment between stakeholders and the engineering team.
  • Designed a mathematical formula for knowledge tracing, estimating students' knowledge states from problem-solving history to address the cold-start problem.
  • Developed rule-based problem recommendation algorithms based on knowledge state levels to deliver problems that meet predefined criteria, using Java and SQL.
  • Designed and implemented database schemas for assessment and recommendation systems in collaboration with the back-end team using Tibero, ensuring essential fields and adding a history-tracking table to prevent duplicate problem recommendations.
  • Developed and deployed RESTful APIs using an in-house framework to deliver recommendation features into production.



Samsung Multicampus

Description

Built personalized recommendation systems using deep learning models, delivering tailored course and content recommendations to over 2 million users.

Contributions

  • Developed deep learning-based personalized recommendation models leveraging sequential user interaction history and user profile data.
  • Developed extensive preprocessing and postprocessing functions to filter out invalid or irrelevant content, ensuring high-quality personalized recommendations in a multi-tenant environment.
  • Built an ML pipeline using Airflow to automate model training and batch inference, enabling continuous recommendation updates via scheduled model refresh.
  • Investigated and resolved operational issues through targeted fixes, ensuring service availability and system stability.

Outcomes

  • Improved user engagement, increasing course enrollment (CVR) by 5% and content completion rate by 25% during the 3-month post-launch monitoring period.



WAPL Math

Description

Conducted a proof-of-concept (PoC) for a deep learning-based Knowledge Tracing model using problem-solving history data to estimate students' knowledge states.

Contributions

  • Achieved an AUC of 88% with a deep learning-based Knowledge Tracing model trained on problem-solving history from 700+ students.
  • Identified limitations in the model’s explainability and proposed directions to enhance interpretability for practical deployment.



Development of Project-Based Peer Learning Service for Workforce Reskilling

Contributions

  • Wrote the AI section of the project proposal and technical documentation, contributing to securing a 3B KRW government project.