Co-teach workshops on Generative AI, specifically fine-tuning LLM models of HuggingFace using LoRA

- Mentor students through online curriculum via 1-on-1 calls, guiding them in developing two capstone projects for their portfolio and advising on career direction- Top 5 most successful mentors across all courses with 75+ student successes and >250 student capstone projects completed

Taught Data Science Introductory course in Python, lecturing and guiding students in completing a machine learning project

Pilotly is a consumer insights platform for the television and film industries that enables content creators to get detailed feedback rapidly from viewers anywhere in the worldDuties/Accomplishments:• Work directly with CEO to determine and implement data science priorities for Pilotly’s online pilot testing platform• Built and maintained Data Science API using Flask to act as Endpoint for key data science products and services and interface with MongoDB and MySQL databases• Built Yelp-style summarizer with sentiment for viewers’ survey responses using noun-phrase counting to find key aspects• Built Selenium based tool to determine if viewers were in the entertainment industry based on Linkedin and Google results• Created sentiment benchmarks by demographic using both survey sentiment and IMDB sentiment data

Idibon uses world-class natural language processing to help multinational organizations find meaning in unstructured text, in dozens of languages.Built ML models to classify text by category and sentiment for ongoing internal and client projects, including: o EDA: Topic Model corpus (LDA), examine distribution, derive insights to determine categories for classification o Annotation: Build labeled datasets by running crowd-sourced annotation studies, evaluating inter-annotatoragreement, accuracy, revising guidelines, iterating o Model Building: Build and Evaluate ML models via cross-validation, build ROC plots, determine optimum modelthresholds and seek to improve via more/better data, or changes to model, feature selection o Synthesis: Visualize model results and insights in Tableau dashboard for ongoing client and internal usePerformed in-depth comparisons of Idibon’s ML platform and sk-learn models (LogisticRegression, MultinomialNB) in Python, discovered bug in feature selection leading to .3 boost in f-score on given data set

• Conducted data analysis and applied machine learning including linear/logistic regression, clustering algorithms, dimensionality reduction, NLP, and time-series analysis using Python’s sk-learn, numpy, pandas, and scipy packages• Built numerous machine learning algorithms in Python, including: random forest classification, logistic regression, k-means clustering• Built data pipelines for scraped data using Python (requests, BeautifulSoup, NLTK), PostgreSQL and MongoDB• Visualized Facebook network and state crime data using: d3, Gephi, Plot.ly, and matplotlibCapstone Project: WeTun.es• Created group playlist recommender that provides multiple playlist options based on individual listening preferences• Built collaborative filter using artist-artist similarity matrix created from the Million Song Dataset’s Taste Profiles – a dataset of user listening history for 1MM users• Clustered artists using affinity propagation for seeding into playlists• Built web application that input Spotify IDs of group members and output playlists ranked in order of predicted group preference along with data visualization of individual playlist preference• Online at wetun.es, code at github.com/bsbell21/weTunes

• Perform data analysis and build dynamic quantitative models to: optimize resource allocation, forecast market movements and long term trends, determine financial feasibility of operational plans, segment customers • Build professional client tools in Excel as deliverables to support strategy implementations• Inform strategy recommendations with qualitative data through client and expert interviews and independent research• Create presentations for use in client meetings, supplier negotiations, and internal knowledge sharing• Selected engagements include: - Coffee and tea worldwide procurement strategy for ~$60B beverage manufacturer - 5-year strategic and operational plan for NYC-based education non-profit - Sales force reorganization for ~$4B beauty products manufacturer

Building and supporting an ecosystem of entrepreneurship in Chile through the development of a network of high-impact entrepreneurs both nationally and internationally• Worked 1-on-1 with entrepreneur candidates, detailing business model, strengths, and weaknesses in 6-10 page profiles used at International Selection Panel for candidate evaluation and critique• Interviewed entrepreneurs, examining wide variety of businesses for high-impact potential• Organized events, speakers, panels, roundtables harnessing Endeavor’s network to support and encourage entrepreneurship• Led proactive search for new entrepreneur prospects in Southern Chile through research, event attendance, and networking

Took ownership of various analyses to drive key insights and contribute to final recommendations for telecommunications and retail clients.

Aided the integration of XO laptops (of One Laptop Per Child) into rural Peruvian classrooms; Consulted to the Peruvian Ministry of Education on the progress and future direction of its OLPC initiative via a detailed report