Code review is one of the essential practices in modern software development and happens very frequently throughout the software life cycle. However, they often incur a lot of cognitive load and require senior competence. Furthermore, some nonfunctional requirements, such as performance and maintainability, are not straightforward to review. Large language models can be used to handle some of the challenges associated with code reviews. However, using large models (LLMs) might incur latency and cost that might not be feasible in many contexts. This paper presents preliminary results of our LLM-based approach for supporting code reviews, which uses program analysis-based methods. The preliminary evaluation of our approach involving user surveys shows promising results. Also, we outline a research roadmap including Retrieval-Augmented Generation (RAG), agentic frameworks, and real-world software engineering integration.