Debugged an NLP-based text analysis app experiencing inconsistent outputs and API latency. Diagnosed tokenization and padding issues for long inputs and implemented preprocessing fixes using Hugging Face Transformers. Improved API performance by addressing synchronous model calls and fixed JSON serialization errors by converting NumPy outputs to native types. Delivered optimized code and clear documentation, enabling stable and fast model predictions under load.