My research focuses on making LLM systems more interoperable, tool-using, and reliable in practice. I build abstractions for unified access across heterogeneous LLM API standards and tool ecosystems, and I study constrained software design and evaluation methods that improve correctness, deployability, and real-world usability.