Matt Legro Email & Phone Number

Marcellus, Ny, Us

Accelerating AI solutions

Evaluate LLM responses for code generation, code comprehension, code debugging, and data analysis and visualization from user-uploaded .csv tasks in Python for accuracy, verbosity, safety, instruction-following and more, providing annotation and feedback for use in LLM performance improvement by fine-tuning

Communicate directly with clients on a Financial Services consulting team, providing NLP, forecasting, and cloud expertise through iterative, Agile-style deliverable development in a fast-paced environment.Project - Voice of the Consumer, Client - CitiBankAn end-to-end NLP and forecasting pipeline with time-series analysis served in a Tableau Dashboard tagged customer complaint feedback with primary complaint intent to inform stakeholders of customer pain points for informing… Show more Communicate directly with clients on a Financial Services consulting team, providing NLP, forecasting, and cloud expertise through iterative, Agile-style deliverable development in a fast-paced environment.Project - Voice of the Consumer, Client - CitiBankAn end-to-end NLP and forecasting pipeline with time-series analysis served in a Tableau Dashboard tagged customer complaint feedback with primary complaint intent to inform stakeholders of customer pain points for informing decision-making on customer service improvement strategy while eliminating manual tagging by service agents. Model development details, using Python:-Identify relevant tables and columns from database table broswer and extract development sets using Hive SQL from Hadoop-based data ecosystem. Ingest data for model training post-POC using PySpark-Generate candidate primary intents using a BERT LLM, fine-tuned on QnA, by prompting the LLM with a set of questions using the transcribed text as context using HuggingFace transformers-Perform semantic clustering on candidates using scikit-learn’s Latent Dirichlet Allocation (LDA) on Sentence-BERT derived to extract exemplar intents and eliminate duplicates, creating an intent pool-Tag future feedback based on a cosine similarity threshold between chunked text feedback and phrases from the intent pool calculated using NumPy, again on Sentence-BERT embeddings-Serve time-series and anomaly-detection insight using an ensemble of rules, moving average, and ARIMA, as well as calculated KPIs like Total Volume and Revenue Loss, in a Tableau DashboardProject - On-Premise to Cloud Migration Strategy, Client - Wells Fargo-Develop a strategy for an on-premise to Cloud transition, architecting a data enterprise solution from stakeholder interviews and iterative design for all levels of the data stack in both GCP and Azure Show less

Remote, United States

Peer review scholarly and technical articles by researchers who do not speak English natively, requiring broad expertise and the ability to process novel experiments and comment on their results. Written review letters lead to publications in journals such as IEEE, Elsevier, ACM, Physica, PRL, and Bioinformatics

Utila, Islas De La Bahía, Honduras

-Teach organism identification to interns in classes and during guided SCUBA identification practice dives-Conduct population health surveys to record long term data-sets in Excel and use data visualization and trend identification in informing local management strategies for Marine Protected Areas in Bay Islands, Honduras

Linthicum Heights, Maryland, United States

Improved electronic chip fabrication process and by increasing the number of Process Control devices measured and analyzed by 10,000 percent through continuous software and test development in MATLAB, fitting statistical distributions to evaluate design metrics and calculate probabilities for production success

Lead groups of secondary school children on multiple day overnight backpacking and canoe-camping trips, teach outdoor skills and guide team-building activities to promote positivity and teamwork in a group environment.

Mentor students in their transition to an independent lifestyle during their formative first year of University and use a positive attitude and communication to develop a healthy communal environment.

Bachelor'S Degree, Physics

Data Science

A five month full time Data Science program covering industry specific methodology and tools, including Python (Pandas, Numpy.

The Modern Python 3 Bootcamp

Interactive series of 30 hours and 200 exercises inlcuding SQL, Web Scraping and HTTP requests