Leading multiple teams responsible for the image processing and machine learning components of SkyWatch's end-to-end delivery to ensure high fidelity satellite imagery to a global client base, delivering more than USD$2M of imagery annually.· Implemented a ML-driven imagery quality solution to identify image artifacts prior to delivery to customers, reducing the need to refund poor quality imagery by > 95%.· Architected and led the implementation of a SotA cloud detection algorithm (custom UNet implementation) to filter potentially obscured imagery and provide cloud masks to interested users (using Tensorflow; see publications).· Rearchitected, and was the primary individual responsible for implementation, of SkyWatch's image processing stack (AWS, Serverless, Python, and using GDAL), reducing the cost to process imagery by 90%.· Invited author to write for AWS Blogs highlighting SkyWatch use cases (AWS EFS, CodeArtifact, CodeBuild, CodePipeline, Lambdas, and Step Functions).· Invited speaker at AWS re:Invent 2023 discussing Edge Computing in space (Las Vegas, USA, 2023).Additional Professional Service· Leading collaborations with University of Waterloo's VIP Lab and University of Toronto's Vector Institute.· Published two peer-reviewed journal articles on machine learning applications on satellite imagery.· Presenter at IEEE WHISPERS 2023.

• Designed and implemented the ML models that formed the core product: Identifying bond trading opportunities for institutional investors: PyTorch implementations of ensemble CNNs integrating static data and multi-featured time series data.• Collaborated in the architecture and implementation of the backend infrastructure: Kubeflow pipelines in GCP, running Apache Beam for ETL.• Oversaw the development, assessment, and interviews of all new technical hires as the team grew from 4 - 20 individuals.• Wrote white papers for fund-raising, participated in media releases, and interviews (e.g., FI-Desk: AI Paradigms for Bond Trading).• Technical liaison in developing partnerships with third-party vendors.

Guided R&D, implementation, and maintenance of the ML models that are replicant.ai’s product: "Faster than real-time" conversational AI (~USD$1.0M annually). Composed of end-of-speech detection, named-entity recognition, intent detection, response prediction (LLMs), and text-to-speech generation. These were implemented in PyTorch using combinations of CNNs, BiLSTM-CRFs, and text-to-Mel spectrogram CNNs (Patent US-20200402497-A1).

• Incepted the design and implemetation of a real-time streaming trading platform for ING’s financial markets’ teams. Backend in Java, Scala, and Python. Co-developed a linear probabilistic model with XGBoost for the real-time price prediction. I was also responsible for the project’s front-end development (HTML5 and CSS, with JS and D3). This product improved returns by upwards of EUR€1M (press releases available online).• Contributed to development of an entity matching algorithm for entities in transactional statements versus ING’s internal nomenclature (implemented in PyCUDA).• Involved in the hiring, mentoring, and supervising of new data scientists into ING’s Wholesale Banking Advanced Analytics, which grew from ~30 to 100+ during my tenure.• Co-supervised two MSc students at the University of Amsterdam.

Contributed to multiple projects across three distinct departments:• Cooperated in the inception of a liquidity/hit-rate recommender system for ING’s Emerging Markets Bonds Desk (in Python); led the development of the recommender system itself.• Implemented a suite of scalable MC simulations that was deployed by ING for calculating default risk charge (in C++ with CUDA); worked closely with the Global Head of Risk Management to classify and quantify intrinsic and systemic counterparty risk.• Implemented a rule-based trade vetting algorithm to assist in automating low volume trades for ING’s bond trading desk (in Java, integrated with ION Marketview Trading Platform).

Conducted independently driven research, developing multi-processor finite-element and -volume hydrodynamical simulations to investigate the mechanisms driving the formation of the first protostellar objects in the universe; utilizing advanced statistical methods to extract potential observational signatures.Thesis: "Observational Signatures from Self-Gravitating Protostellar Disks". (Publications below.)

Worked closely with research scientists and medical professionals in the research and development of a focused ultrasound device for the treatment of individuals with prostate cancer. Directly responsible for the engineering of a focused ultrasound array, and for the development of the image processing algorithms used for extracting thermal deposition information from MRI images.

Assisted with the design, construction, and calibration of a THz time-domain spectrometer. Directly responsible for writing the device control software.

Worked with NASA's MODIS instrument onboard the Terra mission to study the development of extreme weather phenomena. Developed a background in image processing techniques, such as: stacking images to increase S/N; and performing image interpolations to artificially enhance resolution.

Taught Grade 11 physics in a class of 31 students, responsible for lectures, testing, and grading. Additionally developed extracurricular activities for all students, teaching martial-arts classes and starting an ultimate frisbee club.