Welcome
Hello, my name is JooYoung Lee. I am an MS student at POSTECH Data Systems Laboratory, advised by Professor Wook-Shin Han, and I expect to receive my degree in August 2026.
I work on the part where a language model becomes a service: deciding when retrieval is actually needed, checking a draft answer against its evidence before releasing it, and turning natural language into SQL that runs.
My thesis is a negative result. I asked whether LLM-generated consumer personas can forecast demand for products with no sales history, tested it along four routes, and reported that they cannot.
Experiences
- MCP-based ChatRPA ERP (industry-academia project, team). Responsible for the NL2SQL agent and the sales-forecasting module. The target database was undecided when development had to start, so dialect-dependent syntax was isolated behind replaceable points rather than guessed; no rework was needed once PostgreSQL was fixed. Execution was verified against PostgreSQL.
- Event-ticker extraction from economic news (industry-academia pilot). Responsible for the evaluation benchmark and performance measurement.
- LLM unlearning. Found cases where existing unlearning methods stop working once the dataset’s content and format differ from what they were tuned on.
Ran recitation sessions, graded assignments, and held office hours.
Ran the hands-on lab sessions on RAG and MCP for practitioners across several cohorts, following the lecture portion of the curriculum.