Senior AI Scientist

My role at Briteseed was to design Machine Learning models and workflows to enable the detection of critical structures such as blood vessels and nerves in the laparoscopic surgical field. Designed and implemented novel multi-spectral LED control systems and Deep Learning vision models (under significant compute and inference time constraints) to enable real time detection of these critical structures. Control Algorithm Development Designed and implemented proprietary control algorithms that selectively illuminate critical structures in the surgical field. Implemented novel methods to enable the real time detection of structures of interest, and maintain desired intensity profiles across multiple spectral channels, constantly adapting to the changing surgical environment. Implemented efficient caching mechanisms to make use of data across spectral channels to track and recall previously identified structures, improving the decision making frequency of the overall system by more than 15%. Established and manage code repositories for the company's software projects, as well as oversee Briteseed's version control protocols. Deep Learning and Model Development Developed and implemented custom Deep Learning model architectures and training routines on hyperspectral image data to improve Briteseed's real-time tissue identification system. Achieved a >12% increase in critical structure classification accuracy and helped push the software to demo ready. Implemented performant models under strict compute, inference rate and power constraints. Created efficient data pre-processing frameworks to handle large volumes of ex-vivo 3D datapoints. Created visualization boards to evaluate model performance and worked with the clinical team to identify failure modes and accordingly modify ex-vivo data collection procedures. Created model optimization workflows and a code library for efficient iteration, refinement and validation of models. Managed the Deep Learning model stack and created a system to maintain detailed training histories. User Interface (UI) Development Engaged with surgeons and graphic design engineers to establish product requirements and receive iterative surgeon feedback on Briteseed's Surgical user interface. Implemented multiple versions of the interface on the Xilinx ZCU104 development board to overlay real time predictions of the signal processing system on to the surgical camera output.