Natural language processing

• Formed Pickle (pickleai.com), a conversational AI software for outbound sales teams integrated with services such as Zoom, Salesforce, Ring Central, and Outreach.

• As CEO at Convirza created the First-to-Market strategy with an artificial intelligence product that automated natural language processing for an advertising feedback loop into marketing automation, content efficacy, and systems integration.

• Knowledge of how to use Jira, GitHub, Python, Postman, TensorFlow, PyTorch, Google Colab, NLTK, BERT, and spaCy.

Computer vision

• AI Investments and consultation (product, AI, financing support)

• Sentry AI (smartsentry.ai) Computer vision AI applied to millions of security camera videos.

Reinforcement Learning Automated Crypto Trading bot

• My Role: Developed initial model architecture with another engineer for a client trading platform.

• Models: Advantage Actor-Critic (A2C), Reinforcement Learning, LSTM Recurrent Neural Network.

• Libraries: Keras, quantstats, BS4 (Beautiful Soup), NumPy, pandas, matplotlib.

• Tools: Python, PyCharm, Yahoo Finance data sets, Interactive Brokers Integration, Alpaca Integration.

Recommendation engine

• Brand gift recommendation engine: Model provides a recommended brand for a gift recommendation using demographic input provided by the gift giver.

  • My Role: Directed a team of 4 engineers in developing the architecture, strategy, and methods.

  • Models: Evaluated the latest recommendation engine approaches including Collaborative Filtering, Graph Convolutional Network, Graph Attention Networks, Deep Neural Network, Graph Convolution Matrix Completion, and K Nearest Neighbors with various optimizations with hidden layers and embeddings.

  • Libraries: pandas, NumPy, sklearn (MinMaxScaler, Naive_Bayes, NearestNeighbors, matplotlib, RandomForestClassifyer.

  • Tools: Python, PyCharm, AWS Sagemaker, AWS Lambda, AWS S3.

Classification/segmentation

• Inbound Phone Call Intent Classification: Model provided 32 different intent and activity classifications

  • My Role: architect the end-to-end model. Ran daily scrum meetings and coordinated with product and QA. Also identified and implemented Automatic Speech Recognition.

  • Models: Transformer models using Bert, RoBert and ran initial tests on GPT-3, Convolutional Neural Networks with significant adjustments for data imbalance.

  • Libraries: scikit-learn, pandas, NumPy, NLTK, glove, matplotlib.

  • Tools: Python, Tensorflow, PyCharm, AWS Sagemaker, AWS Lambda.

• Lead quality scoring and sale prediction: Model provides a lead score a recommended action (e.g., callback)

  • My Role: Integrated dataset and led project with 2 engineers in developing the architecture, strategy, and methods.

  • Models: Multivariate regression model initially and then added natural language data to measure intent using Convolutional Neural Network with GloVe (Global Vector of word embeddings)

  • Libraries: pandas, NumPy, NLTK, glove, matplotlib.

  • Tools: Python, Tensorflow, Google Data Studios, Google Colab.