Providing AI expertise to companies from pre-seed to growth stage and beyond. Building and maintaining Moonshine's open source tools.

Researched the next-generation self-driving ML systems. Foundation models for AV perception, end-to-end learning for perception and behavior, multimodal LLMs.

Built robust perception systems for trucking. State of the art lidar and camera based deep learning, foundation models for vision, domain translation from surface streets to freeway. Tech lead nominal perception, large vehicles, trucking lidar segmentation, freeway data & evaluation.

Led a 10 person team of researchers and engineers focused on improving object detection at Cruise. Built, maintained, and iterated a primary production object detector. Conducted forward looking applied research in object detection, tracking, and simulation. Reduced object detection related safety events by 40% in 6 months.

Tech lead and project owner: emergency vehicles, audio understanding, lidar detection (birds eye view)Developing and deploying machine learning systems for perception. Deep learning for computer vision, audio, lidar, and sensor fusion. Responsibilities include or have included:- Model development and experimentation- Data curation and pipelining- ML infrastructure- Measurement and evaluation- Hardware definition and validation

Spearheaded effort to introduce deep learning to Fitbit R&D. Developed and validated Tensorflow-based networks on a variety of data modalities, including building associated training infrastructure. Built machine learning tooling for model deployment both to the cloud as well as low-power micro-controllers. Researched machine learning and signal processing methods for a variety of data, including images, accelerometer/gyroscope, PPG, and audio. Wrote firmware that shipped on products, including on the smartwatch Fitbit Ionic and Versa. Acted as a technical mentor and leader, presenting at internal talks and managing/mentoring interns.

Advisor: Zeeshan SyedInvestigated methods for predicting cardiac complications by applying convolutional to sensor data such as ECG. Developed data pipelines and applied statistical methods to segregate patients based on predicted postoperative atrial fibrillation risk. Built and open sourced Mri, a web based tool for monitoring and validating deep learning models.Researched advanced methods of static compiler optimization by applying deep sequence models to LLVM intermediate language.

Researched time and spectral pitch detection algorithms for audio pitch correction software. Implemented and evaluated pitch algorithm candidates and realized 10% improvement from current method, deployed in Nectar 2 and other commercial software.

Began internship at age 16.Designed and deployed commercial intercom control software suite. Developed hardware environment sensors and associated control algorithms. Researched and designed an active noise canceling system for industrial blowers.