• Led development of comprehensive AI recipe personalization engine integrating OpenAI across multiple recommendation contexts: dynamic recipe prompt suggestions, holiday and cookbook-based inspiration, pantry-aware meal planning, and user likes/dislikes

• Worked directly with customers to address product feedback, delivering bug fixes and feature improvements with same-day turnarounds

• Built agentic cookbook recognition system with OpenAI GPT-4.1 and intelligent ISBN matching, transforming personal cookbook libraries into digital recipe sources

• Created real-time messaging portal with thread-based conversations, implementing optimistic UI updates with React Query, anonymous authentication, and multi-asset attachment support

• Designed reusable component library with 40+ components using Tailwind CSS v4 and Catalyst UI design system

• Implemented Simplex Volume Maximization with Python to create user archetypes, enhancing user segmentation and targeting strategies.

• Developed EOD revenue prediction models using BigQuery and Python, and optimized with hyperparameter tuning.

• Developed flexible Looker models and dashboards, enabling the content team to optimize game difficulty, contributing to more consistent revenue streams.

• Created Live Ops revenue attribution pipeline to optimize LiveOps event types and IAP offerings.

• Optimized and migrated ETL scripts from Vertica to BigQuery using Airflow, and implemented Python-based automated monitoring for database migration integrity.

• Automated key reporting processes, including an executive revenue overview tracking $250M, enhancing report accuracy and efficiency.

• Highly skilled in creating, optimizing, and maintaining complex data pipelines for real-time, historical, and predictive usecases

• Created custom, interactive visualizations and dashboards for many different industries, users, and scopes

• Built AI models by cleaning and engineering features from customer Salesforce data

• Productionalized AI predictions by building custom web components

• Ramped quickly with each new company/client (10+) on projects lasting 1-12 months

• Communicated directly with client stakeholders and technical leads to define business requirements

• Estimated, planned, and prioritized multi-phase projects for longer engagements (12+ months)

• Certified "Tableau CRM & Discovery Consultant" and "Platform Developer 1"

Learning to solve machine learning problems in a professional development environment with the help of a personally assigned mentor. Leveraging previous experience in machine learning, Python, Pandas, Numpy, and Matplotlib to learn the statistics of data science and complete projects in supervised & unsupervised learning. Implementing state of the art libraries such as TensorFlow and Keras in the fields of deep learning, NLP, and Computer Vision.

Projects:

Seq2Seq Chatbot (coming soon to my GitHub page, post-bootcamp):

The chatbot was trained on over 30,000 text messages between myself and my girlfriend. The model was written in Keras and uses a GAN autoencoder. (work in progress....)

Contract Cheating Classifier

Trained on data scraped from various contract cheating and student essay repository sites. Several Scikit-learn algorithms were trained and validated using a custom random search class to tune hyperparameters. The best algorithm was chosen among various SVC kernels, Random Forest, K-Nearest Neighbors, as well as other ensembles (bagged, boosted). Dimensionality reduction techniques used were LSA, LDA, NNMF, and Gensim's Doc2Vec. Human equivalent f1 score of 0.62 was acheived.

IMDB Movie Score Classifier:

Trained on comments scraped from YouTube movie trailers. Trailer sentiments were used to predict if a movie scored above or below a 6.4 on IMDb's website. Random Forests, SVM, K-Nearest Neighbors, and other ensemble methods were tuned and evaluated on an f1 metric, from which the best algorithm was chosen. Best f1 score of about 0.71 was achieved.

Classic Literature Author Predictor:

Given a sample of text from classic books such as Alice in Wonderland and Moby Dick, a Linaear SVC algorithm was able to obtain a accuracy of 0.96 on 126 authors. Unupervised dimensionality reduction techniques (TFIDF and LSA) were used as well.

GitHub link: https://github.com/bgwilson12

Designed and drafted HVAC, plumbing, and medical gas systems for construction of pharmacies, hospitals, and commercial buildings. Communicated with architects, electricians, contractors, and other engineering disciplines to deliver more than 15 large scale project designs on time.