Mudasir Syed

Multimodal AI and vision-language models, with a focus on fine-tuning (LoRA/PEFT, GRPO) for structured output generation — from floor-plan vectorization to conversational query understanding. Interested in retrieval-augmented generation (RAG), context-aware NLP for conversational search, and building practical instruction-tuning datasets that bridge ambiguous human language with structured system commands. Currently exploring reinforcement learning methods (GRPO, geometric/structured rewards) for improving VLM output reliability.