Experience
TmaxRG
Researcher, Diagnosis and Recommendation Team
Jan 2023 – Nov 2024
As a Researcher on the Diagnosis and Recommendation Team at TmaxRG, I built models that decide which course or problem a learner should see next.
I developed recommenders, improved knowledge-tracing models, and built the Airflow pipelines that deliver results to clients. Through client meetings and close work with the team, I helped turn learning data into recommendations delivered to client sites.
01. Course recommendation
One of my primary focus areas was course recommendation for Samsung Multicampus, helping learners find courses that fit their learning history.
- Raised enrollment conversion by 5% and completion by 25% over the previous recommender by developing a course recommender that uses learning histories and BERT-based title similarity.
02. Learner diagnosis & problem recommendation
I also worked on diagnosing what learners know and recommending problems that match, helping digital mathematics textbooks adapt to each student.
- Improved knowledge-tracing models for learner diagnosis by engineering new features from WAPL Math, an AI-based mathematics learning platform.
- Led development of a problem recommender tailored to student proficiency for digital mathematics textbooks, gathering requirements in client meetings.
03. MLOps
I built the pipelines that keep recommendations current and deliver them to clients.
- Built Airflow pipelines for retraining and batch inference, writing per-user recommendations to the databases of about 800 client sites.
04. Additional contributions
In addition to modeling, I contributed to proposals, mentoring, and company-wide recognition.
- Participated in writing and presenting a KEIT research proposal on peer learning for worker reskilling, funded with KRW 2.7 billion in government grants over three years.
- Mentored a new hire through onboarding in the New-Hire Mentoring Program (Feb – May 2024).
- Awarded the Super Rookie Award at Tmax A&C (2024), given to outstanding first-year employees across its affiliates.