Ali Hashemi

AI researcher specialized in optimizing foundation models and LLMs for efficiency, interpretability, and scalable deployment. Experienced in model distillation, representation learning, and generative modeling for biomedical and scientific domains. Highlights: - Extensive expertise in model distillation, sparse autoencoders, quantization, pruning, and parameter-efficient fine-tuning (PEFT). - Developed innovative distillation methods reducing model size and latency by up to 80% without accuracy loss. - Strong theoretical foundation in generative models (diffusion, consistency), Bayesian statistics, and geometric/non-convex optimization, with 10+ top-tier publications (including NeurIPS). - Applied representation learning to computational histopathology and quantum chemistry, enhancing accuracy and boosting efficiency. - Proficient software engineer skilled in Python, PyTorch, Docker, Git, AWS, GCP, CI/CD, distributed training, and model serving (TensorRT, ONNX, vLLM). - Successfully led international projects delivering impactful AI solutions in biomedical and multimodal domains. My research focuses on optimizing foundation models and LLMs using advanced representation learning methods, emphasizing efficiency, interpretability, and scalable inference. Techniques include model distillation, sparse autoencoders, sparsity-induced priors, quantization, pruning, and PEFT. My theoretical expertise covers generative modeling (diffusion, consistency), Bayesian statistics, graphical models, and geometric/non-convex optimization, demonstrated through multiple publications at prestigious conferences. Currently, I apply these methods to optimize foundation models via white-box distillation, significantly reducing model size and latency. This work directly impacts fields like computational histopathology and quantum chemistry. Previously, I developed probabilistic graphical models and inference algorithms for neuroimaging, employing sparse Bayesian learning and geometric optimization. These innovations improved analysis precision for brain mapping, brain-computer interfaces (BCIs), and E/MEG reconstruction. As an experienced software engineer, I specialize in Python, PyTorch, Docker, Git, AWS (SageMaker, EC2, S3), GCP, CI/CD pipelines, distributed training, and serving models using TensorRT, ONNX, and vLLM. My engineering has significantly improved efficiency in healthcare, scientific computing, and multimodal learning. Throughout my career, I have successfully led international projects, bridging theoretical innovation with practical, scalable AI deployment.