SUMMARY:

• My goal is to obtain employment as a lead or principal machine learning engineer, lead or principal data scientist, data science manager or senior manager, chief data scientist or chief data officer, or similar role.
• Experienced: Neural networks, support vector machines, deep learning, Confidential, LSTM, RNN, BERT, GPT, GPT - 2, GPT-3, NLP, RL, Reinforcement Learning, IOT, Internet of things, transformer models, Soft Actor Critic, SAC, CNN, Capsule Networks, attention networks, gradient propagation algorithms, Tensorflow, Keras, PyTorch, dense layers, convolutional layers, Multivariate analysis, (principal component analysis (PCA), non-negative matrix factorization(NNMF), parallel factors/ canonical decomposition (PARAFAC/CANDECOMP), Tucker decompositions, Poisson regression), data clustering (K-means, K-medoid, DBSCAN, Shared Nearest Neighbors (SNN), K-Nearest Neighbors (KNN) hierarchical/agglomerative, distance metric selection, finding number of clusters (Davies-Bouldin, Calinski-Harabasz, silhouette)), data munging (low level text extraction scripts, missing values, template matching), visualization (Matlab, Matlibplot, Graphical User Interface (GUI), custom display)
• Intermediate: Ensemble methods (random forest, KNN subspace learners, feature transformation), feature selection (F-test, unsupervised/supervised methods, iterative methods), model/curve fitting (linear/logistical regression, splines)
• Experienced: Tensor/matrix factorization math, optimization algorithms (interior point, trust region, gradient decent, grid search), global minimum searches, sensitivity measurements (initial condition perturbation, error tracing), simulation methods (finite difference time domain (FDTD), parameter estimation),
• Previous Experience: Statistics (Bayesian, frequentist, power, F-test, T-test, probability distributions, ANOVA, A/B testing)
• Software Development Consultant for G86 Studios, Salt Lake City, Utah, 2012
• Small startup company specializing in industrial, graphic, user interface, and video game design
• Consulted on a project to develop Windows 8 Tablet games using C# in the .Net framework, specifically using the XNA Game Developer Studio Libraries
• Working knowledge of C# programming using MSDN as a result of the experience

TECHNICAL SKILLS:

Experienced: Matlab language (includes designing, writing, debugging, and deploying GC-MS data processing GUI with 10,000+ lines of code in interface/libraries), Python, Python libraries (Numpy, SciPy, Scikit-learn, Matplotlib, h5py, f2py, ctypes, PyQT, Pandas), Python machine learning platforms (PyTorch, Tensorflow, Keras, Onyx), R, SQL

Previous experience: Html5, C#, Java, R, BASH scripts, C++, C, BASIC, Visual Basic

Previously Used Code Development Environments: Matlab IDE, Anaconda Spyder, Microsoft Visual Studios, BASH, Vim, Eclipse, Liclipse, Notepad++, Git, Gitlab, Jupyter, AWS, Domino

Data Formats: HDF5, JPG, JPG 2000, compressed file formats, JSON, YAML, Markdown, Neo4j

WORK EXPERIENCE:

Vice President, Lead Developer

Confidential, Greenwich, Connecticut

Responsibilities:

• Lead a team of software engineers, UI/UX engineers, and data scientists to improve Confidential 's intelligent document processing capabilities using AI/ML tools, including OCR, NLP, GCP, etc. Developed expertise with and led implementation of new features and machine learning algorithms in a tech stack that combined modern website languages and tools (HTML, CSS, JavaScript, ReactJS, Ruby - on-Rails, Docker, Kubernetes) with bleeding edge machine learning tools (Python, Jupyter Notebooks, AutoML, PyTorch, VertexAI, BERT Transformers).
• Lead hiring initiative to hire and internally recruit most members of my current data science team. This includes two junior data scientists, a mid-level data scientist, a senior data science product manager, two full time software developers, and a web designer. I was responsible for leading and designing the interview process, including performing an initial technical screen to assess capabilities and implementing a coding exercise that I designed that tested a candidate’s understanding of basic data science tools. I am currently leading an effort to hire three more people into the team.
• Designed, built, and implemented most of the current company MLOps stack and process. This includes writing specific Bash and Python files that utilize the Google Cloud SDK to automate dataset creation, model training and deployment, model assessment metrics, and data governance. I also assembled multiple datasets for model training, directed labelling and curating of the datasets, and established metrics to measure dataset quality.
• Consulted regularly on machine learning with business product owners, development teams, and external investors about strategies to bring value to the company through AI/ML utilization. I report directly into the C-Suite and have had several conversations with and presentations to leading company officers, including the Chief Technology Officer and the Chief Data Officer.

Staff Software Engineer

Confidential, Shelton, Connecticut

Responsibilities:

• Designed, wrote, debugged, and demonstrated to customers a game inspired by Battleship that uses reinforcement learning to help train AI agents to efficiently allocate resources on a game board.
• The game uses an OpenAI interface and is packaged for Pip installation.
• I also created and trained an AI controller using a custom model that I created in Keras and Tensorflow 2.0.
• The custom model uses several features that allow it to account for changes in the game environment, including changes to the game board size and number of player-controlled assets.
• Implemented a Recurrent Neural Network (RNN) on the same data to leverage time series nature of state X action space.
• Mentored junior data scientist seeking to implement a Monte Carlo Counterfactual Regret Minimization (MCCFR) algorithm in the same problem space.
• Assisted in development of a pipeline for processing Internet of Things (IOT) data.
• My work included helping to clean the database access and munging code, writing unit tests, and documenting features via docstrings and Jupyter Notebook tutorials.
• I also added several new features to the code base, including upgrading the code to Tensorflow 2.0.
• Hosted training session where I showed Jupyter notebooks to customers and internal data science teams. Also functioned as a consultant for several internal data teams seeking to implement IOT solutions to add value.
• Worked on Natural Language Processing (NLP) project that sought to map error reports to a classification graph tree embedded in n-dimensional hyperbolic space.
• Gained experience with Bidirectional Long Short-Term Memory (BiLSTM) models and BERT and associated (RoBERTa, GPT-2) transformers.
• Experience included fine-tuning BERT model using custom text corpus.
• Worked on project that investigated using reinforcement learning (RL) to train agent to perform tasks in vehicle simulation environment.
• Implemented both a custom version environment and a customized version the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm for the problem.
• Programming was done in PyTorch and native Python. Skills obtained included broad examination of RL literature, setting up cloud computing (AWS) environments using Docker and web interfaces, and executing directed hyperparameter searches using Bayesian techniques. The results showed the method was promising, including monotonic convergence of the custom loss function and demonstrated learning behavior in restricted environments.

Data Scientist, Senior Member Technical Staff

Confidential

Responsibilities:

• Lead a team of researchers that implemented a Capsule Network in Keras to classify satellite image chips. Achieved 40% identification success rate versus 20% success rate from neural network tested in dataset reference paper (which used a You Only Look Once (YOLO) Single Shot Detector (SSD)).
• Lead a team of researchers that implemented a Generative Adversarial Network in PyTorch to create high quality synthetic images of synthetic aperture radar (SAR) data. The team employed advanced concepts, including a sliced Wasserstein loss function and the Pix2Pix image translation package. A test asking humans to discriminate between real images from the training set and fake images produced by the GAN was able to fool non-experts as if they were guessing randomly and experts at a statistically significant rate.
• Leveraged tensor factorization to classify high energy events from 1-dimensional spectral datasets. Also employed tensor completion algorithms to recover datasets with up to 90% of the entries missing (due to salt and pepper noise) with high fidelity. I researched, wrote, debugged, and deployed 100% of the source code for two recovery algorithms studied in the project.
• Used matrix and tensor factorization methods to recover pure spectra from chemometric data (including Gas Chromatography Mass Spectrometry data) of outgassing samples from field systems.
• Built a commercially deployed GUI and library (written in Matlab) that automated chemical identifications using callbacks to commercial spectral databases. I used advanced statistical tools, including Varimax rotation and Poisson noise estimators, to increase identification accuracy. The software was fully deployed in the form of an installation package to customer computers, which resulted in large improvements in process efficiency and money savings.
• Built and deployed a Bayesian network (using the bnlearn toolbox in R) for waveform identification and event classification.
• Deployment included network construction using both expert elicitation and iterative methods from an input dataset. I also used Markov Chain Monte Carlo (MCMC) methods for simulated system parameter estimation.
• Translated a 10,000+ line program that was designed to simulate the operation of a piece of industrial equipment from a proprietary programming language into Python 2.7. The translation included writing custom library calls with Cython and F2PY, translating equations using the SymPy libraries, and designing and writing a custom GUI interface using the TKinter library.
• Have written over 100K lines of Python and Matlab code and several thousand lines of R code. I have used several major toolboxes in all three languages, including most major toolboxes in the Python language (Numpy, Scipy, Scikit-learn, Pandas, Sympy, etc.).
• Have written reports for customers using Jupyter Notebooks, PowerPoint, LiveScripts (Matlab), Latex, and Microsoft Word and am comfortable presenting results in all formats.
• Tested algorithms for compression of Synthetic Aperture Radar data, including discrete cosine transformations and discrete wavelet transformations.
• Examined various compression algorithms, including Set Partitioning in Hierarchical Trees (SPIHT) and error correction algorithms
• Experienced with concepts like Discrete Fourier Transforms (DFTs), noise removal, signal preprocessing, filtering, and domain conversions

Data Scientist, Postdoctoral Appointee

Confidential, Albuquerque, New Mexico

Responsibilities:

• Wrote 5,000+ lines of Matlab code to develop, debug, and deploy a robust, user-friendly Graphical User Interface (GUI) to import, analyze, and visualize large chemometric data sets.
• Gained extensive experience with data space reduction methods, including principal component analysis (PCA), multivariate analysis, matrix factorization, tensor decomposition and compensation for Poisson noise
• Developed, analyzed, refined, and implemented algorithms for automated parsing and analysis of large data sets, including an algorithm to determine the rank of a factored data matrix
• Developed robust simulation code in the Matlab language to produce realistic simulated elution profiles and mass spectra to isolate noise effects and analyze algorithm efficiency

Adjunct Professor

Confidential, Salt Lake City, Utah

Responsibilities:

• Presented material at a level which was sufficiently rigorous for formal mathematics courses but which could be easily digested by students
• Prepared extensive lecture notes, exams, and supplemental material to aid in teaching the class
• Attended training sessions for the E-Portfolio initiative - an innovative program that helps students to develop an online resume of projects completed during coursework