Technical lead overseeing multiple projects developing VLMs (vision-language models) to support SambaNova’s generative AI ofering (utilizing models like CLIP, LLaVA, mLlama3.2). My responsibilities include working closely with our compiler team to optimize model performance, collaborating with the production team to ensure enterprise-scale deployment, and verifying our ML solution efectively addresses the customer's business needs.

Managing the computer vision and multimodality team for SambaNova's Foundation model offerings

• Working on Automatic Speech Recognition Systems. Including bring up of HuBERT on SambaNova's custom hardware and providing fine tuning capabilities for customers.• Leading efforts to build large transformer-based computer vision models (e.g. SWIN) that can take advantage of true high-resolution imagery (e.g. 6k imagery in medical and satellite applications)

• Co-developed a deep learning model to accelerate scientific simulations of multi-physics Inertial Confinement Fusion (ICF) experiments at the National Ignition Facility. Accelerated simulations of X-ray spectroscopic signatures by > 100x• Built MCMC inference routines utilizing the deep learning accelerated X-ray simulations to solve the inverse problem and obtain estimates of unobservable parameters in experiments (e.g. temperature and density)• Providing technical guidance on implementing Bayesian Neural Networks and unsupervised learning to interpret experimental data at the National Ignition Facility

• Managed a team of six people to take an experimental codebase to production. Put in place best practices, CI/CD, and worked with customer-facing teams to improve workflows• Managed high-risk R&D projects on the critical path. Worked directly with the senior leadership team to plan roll-out of computer vision applications to early-access customers using agile methodologies• Lead engineer overseeing neural network bring-up efforts for applications like pose estimation, monocular depth estimation, and object detection. Worked with teams across Mythic's technical stack to adapt customer networks to Mythic's hardware

• Focused on the co-design of neural network architectures and domain specific hardware to accelerate computer vision inference applications• Worked closely with digital and analog circuit design teams to develop software simulations of hardware nonidealities. Utilized these simulations and quantization-aware training methods to generate networks that were robust in mixed-signal and analog environments

• Worked as an applied researcher to rapidly-prototype proof-of-principle deep learning applications in computer vision - showcasing Mythic's capabilities for early customer engagement• Organized internal and external teams as a project manager to build deployment pipelines for edge use. Built plugins for GStreamer to run pre- and post-processing code that interfaced with Mythic's custom hardware

• Providing mentorship for research and engineering projects across a variety of applications in the deep learning space: Computer vision, Deep Reinforcement Learning, Generative Adversarial Networks

• Consulted for Harvesting Inc., focused on leveraging AI and remote-sensing to assist farmers in rural areas and developing countries• Engineered and implemented a deep neural network for object detection and identification in high-resolution satellite images• Applied techniques in transfer learning and data augmentation to achieve high-performance despite limited data

• Worked with a small team to develop a high-intensity laser system for a first-of-its-kind research facility• Performed research and spearheaded rapid-prototyping projects encompassing multiple engineering disciplines: electrical engineering, mechanical engineering, chemistry, and computer programming• Developed routines using Python for error analysis, interfacing with commercially-available software, and image processing to streamline the work of colleaguesTechnical Summary: Explored the deviation of experimental measurements from the theoretical linear mixing model of hydrocarbons under shock compression to pressures at 4 Mbar.

• Organized interdisciplinary teams (~5 people) on a biannual basis to complete short-term (~2 week), high-value (>$100,000), projects to support the interests of the Department of Energy and National Nuclear Security Agency• Built and fielded highly-technical diagnostic systems (optical, X-ray, and nuclear) to explore fundamental questions in high-energy-density physics• Performed physics simulations, using massively-parallel computing platforms, to analyze and interpret experimental resultsTechnical Summary: Investigated instability-driven relativistic (~ 1MeV) electron beams in X-pinch pulsed power experiments

• Designed and built scientific equipment and diagnostic systems - used at Lawrence Livermore National Laboratory, Argonne National Laboratory, and the Stanford Linear Accelerator• Performed experiments studying materials under high-pressure (> 1 Million Atmospheres), resulting in several high-impact publicationsTechnical Summary: Built and fielded laser-interferometry diagnostic systems for high-velocity (> 10 km/s) shock wave measurments

• Assisted with development of COFE, the Cosmic Foreground Explorer Telescope, a balloon-borne telescope with the mission to study the polarization spectrum of the Cosmic Microwave Background