Joint Research Project on Causal Inference with a Research Institute (LLM), Oct. 2024~Dec. 2025

We conducted a Proof of Concept (PoC) on causal inference over healthcare-scale datasets using Large Language Models (LLMs), with emphasis on localized and domain-specific deployment under data confidentiality constraints.

While project objectives were predefined, the formalization and sharpening of the research program were carried out primarily under my responsibility. A central axis was to analyze the contribution of item-intrinsic signals—particularly temporality—to causal effect estimation under intervention, and to align heterogeneous estimators under a common evaluative framework. The resulting prototype integrated multiple causal inference methods and supported comparative analysis across modeling assumptions.

My role combined research design and project management. On the research side, I specified the problem at an operational level: selecting relevant substructures from high-dimensional data, defining method-consistent ground truth, and establishing evaluation criteria across estimators. On the execution side, I designed the experimental pipeline with explicit requirements for traceability and reproducibility, coordinated task allocation among AI engineers, and maintained structured iteration through regular technical reviews.

I was also responsible for consolidating and finalizing artifacts, including formal responses to external review by the research institute. The PoC produced structurally interpretable results, informing both methodological questions in causal inference and domain-specific considerations, and provided a basis for subsequent research development.