Developed a model forward pass profiling tool for the unslothai/unsloth repository, focusing on enabling kernel-level optimizations in machine learning workflows. The tool was designed to surface components within the model that are suitable for kernelization, providing actionable insights for targeted performance improvements. Leveraging Python and PyTorch, the developer implemented profiling instrumentation and established a data-driven workflow to guide optimization decisions. This work improved visibility into inference bottlenecks and laid the groundwork for enhanced throughput and reduced latency. No major bugs were addressed during this period, with efforts concentrated on performance profiling, model optimization, and cross-repository collaboration for business impact.