Contributed to the modular/modular repository by expanding model integration capabilities within the MAX framework. Developed support for embedding tasks using BertModel with sentence-transformers/all-MiniLM-L6-v2, enabling broader NLP experimentation without strict CUDA dependencies. Added architecture registration and configuration for non-MoE Qwen3VL 2B and 4B variants, supporting diverse text generation workflows. Collaborated through pull requests, co-authoring architecture changes and closing integration issues. Focused on Python-based deep learning and model deployment, the work emphasized modularity and benchmarking to guide future optimization. No major bugs were reported, reflecting a focus on feature delivery and robust engineering practices in machine learning model development.