Raushan Turganbay

- Worked on shipping LVLMs such as Chameleon, LLaVa-Onevision, Mllama etc. - General maintaining of `transformers` library and API standardization for LVLMs - Plan to optimize the performance of video LLMs

- Shipped a few multimodal generative models, including Video-LLaVa and LLaVa-NeXT-Video. - Implemented KV cache quantization for optimizing memory usage in large-language models. - Integrated watermarking technique that works by adding a bias to certain tokens with a private hash key - Enhanced model performance and efficiency by enabling `torch.compile` compatibility for a set of LLMs

- Worked on rule-based intent classification. - Performed user testing to ensure optimal performance of the conversational AI

Facilitated interactive language learning experience and promoted cultural understanding among students in a summer school program.

Research topics: acoustic cues of stress, multilingualism, linguistic rhythm