Bridge Neurotech is building a non-invasive functional ultrasound imaging (fUSI) platform with AI-based neural decoding for brain-computer interface applications. It is a seed-stage neurotechnology company based in Redwood City, with custom hardware, imaging science, real-time compute, neural decoding and research studies under one roof.

About the role

This role owns and scales key parts of the models and infrastructure behind Bridge's neural decoding pipeline, from raw functional ultrasound acquisition through image reconstruction, preprocessing and decoding. It reports to the ML/AI Engineering Lead.

Recent work has shown brain activity can be mapped into the latent spaces of pretrained deep learning models to reconstruct visual perception and continuous language, demonstrated with fMRI in research settings. The job is to bring those techniques to functional ultrasound, a modality with better spatial and temporal resolution, portability, and a path to consumer deployment.

What you will do

• Design, implement, train and evaluate models across fUSI reconstruction, signal preprocessing and downstream decoding.
• Drive the experimental agenda: test hypotheses about why a model is not learning, design ablations, and tune architectures, losses and training schedules.
• Extend models to new data regimes, including 2D to 3D volumetric data, new probes and new subjects.
• Own the training infrastructure: experiment tracking, config management, checkpointing, mixed-precision and multi-GPU training, and profiling.
• Define the evaluation strategy, and build inference paths for full-volume reconstruction and low-latency real-time decoding.
• Work with the imaging scientists and neuroscientists to understand where the data comes from and what its artifacts mean.

What they are looking for

• A PhD in machine learning, computer science, electrical engineering, computational neuroscience or physics, or a bachelor's or master's plus 2 to 3 years building and shipping ML systems.
• Strong Python and hands-on PyTorch, JAX or TensorFlow development.
• A track record of taking a model from paper or prototype to something that trains reliably on real data.
• Comfort with large numerical datasets and memory-mapped arrays, and GPU proficiency across mixed precision and memory profiling.
• Nice to have: neural operators, implicit neural representations, physics-informed architectures, or exposure to neural decoding and neuroimaging data.