Machine learning (ML) is widely used across domains, with Jupyter Notebooks serving as a key platform for ML prototyping. Enhancing code quality in ML notebooks is essential but challenging due to dynamic typing, complex ML libraries, and notebook-specific semantics. Static analysis detects bugs without execution, offering fast and early feedback during development. However, it often fails in ML settings where types and behaviors are not statically known. Large language models (LLMs) show promise for code understanding and bug detection, yet suffer from issues like hallucination and lack of execution grounding. We propose semi-static analysis: a hybrid approach that combines static analysis, LLM, and runtime information from executed notebook cells, to improve static bug detection in ML notebook development.