Trevor Dalton

About Me

Senior Data Engineer at M Science, where my focus is now on agentic AI, building pipelines that automate large-scale data labeling and classification for analyst consumption. Passionate about LLMs and the engineering challenges of putting them into production.

When I'm not coding I'm lifting weights, playing pickleball, or reading. Currently on The Count of Monte Cristo by Alexandre Dumas. Always happy to talk books or anything else.

Source Code

Experience

M Science

Senior Data Engineer

Period: January 2026 - Present

Data Engineer

Period: March 2022 - December 2025

Designed an agentic eReceipt tagging pipeline using LangGraph, BM25 retrieval, and OpenAI/Anthropic LLMs to automatically classify thousands of daily transactions against a video game product taxonomy
Fine-tuned multiple BERT-based classification models using PyTorch and Hugging Face to automate large-scale data labeling and improve data accuracy with F1-scores above 98%. Deployed models with MLflow, streamlining operations and cutting manual effort by 50%
Derived actionable insights from large-scale video game sales data using SQL and Python, uncovering user behavior trends to inform client product strategy and marketing decisions
Construct PySpark/SQL ETL pipelines for full raw-to-deliverable processing of large-scale data
Optimize AWS EC2 configurations to cut pipeline costs and runtime by as much as 60%
Developed DataOps-controlled pipeline orchestration systems, reducing runtimes by over 20%

Python

Databricks

Apache Spark

PyTorch

Hugging Face

Snowflake

AWS

Apache Airflow

MySQL

Git

LangChain

Anthropic

Northrop Grumman

Data Engineer

Period: May 2021 - March 2022

Data Engineering Intern

Period: May 2020 - May 2021

Engineered a JavaScript multi-source ETL pipeline to connect, aggregate & analyze data
Designed frontend interface for a master data pipeline to derive insights from enterprise application data
Developed an Ontology-driven ETL Orchestration tool used by 50+ developers using React
Built scalable desktop applications using React, Node.js, TypeScript, and Electron
Developed machine learning and graph database models for supply chain and maintenance forecasting to mitigate risk and reduce lifetime costs of advanced weapons systems
Delivered valuable insights on large datasets using statistical analysis and MatPlotLib visualizations

Python

TypeScript

JavaScript

Node.js

React

Neo4j

Apollo

GraphQL

mongoDB

MySQL

AWS

Git

Education

University of California at Berkeley

Master of Information and Data Science

The MIDS program at the Berkeley School of Information is recognized as one of the nation's top-tier data science program. The program's focus on collaborative problem-solving has taught me how to form effective teams from a diverse set of individuals.

Graduated In: 2024

Natural Language Processing
Time Series and Panel Data Analysis
Computer Vision
Experiments and Causal Inference
Machine Learning
Data Engineering
Statistics
Research Design and Analysis

University of Utah

Bachelor of Computer Science

Utah's preeminent research institution and where I first cut my teeth at Software Development. Despite the challenging curriculum I was able to thrive thanks to the help of my peers and professors who offered ample support.

Graduated In: 2021

Algorithms
Artificial Intelligence
Data Visualization
Operating Systems
Database Systems
Machine Learning
NLP
Information Systems
Computer Systems

Utah Tech University

Associate of Science

Taking full advantage of the Success Academy Program I engaged in concurrent enrollment classes which allowed me to graduate high school with enough credits to attain by associates degree from Utah Tech University at the age of 18. The supportive community and passionate professors ignited my love for computing that continues to today.

Graduated In: 2018

Data Structures
Algorithms

Projects

Datadrip: AI for Financial Analysts

Datadrip automates earnings analyses for publicly traded companies using state-of-the-art technology. It frees research analysts from manual work by extracting and summarizing critical data instantly. Retail investors gain rapid insights into revenue trends and presentation summaries in a concise format. For financial analysts, Datadrip converts presentations into Excel sheets for seamless modeling integration, enhancing decision-making and maximizing returns effortlessly.

During Datadrip's 16-week development, the Datadrip team leveraged cutting-edge models in computer vision, generative AI, and visual Q/A. Datadrip exemplifies the evolving landscape of AI-driven financial analysis tools, highlighting the potential for cutting-edge technologies to democratize access to critical financial information and elevate data-driven decision-making in the financial sector.

You can view the demo here!

Website

Bank Document Verifier

A collaborative workspace geared to enable financial institutions to automate their application processes. Created using Angular, TypeScript, and SCSS. Here, banks can automatically verify important documents such as W2s, Schedule C's, etc. and approve applicants. BDV allows for real-time communication to ensure loans are processed accurately and timely.

The Bank Document Verifier team was awarded 2nd place in the University of Utah's School of Computing's 2021 Spring capstone presentations

Demo

Aye-Aye: Semantic Lexicon Induction

A semi-supervised lexicon induction algorithm! Given just a few seed words this the Aye-Aye can learn any semantic category to a high degree of precision

Github Repo

Personal Website

Passion project of mine I have maintained since 2018 to keep with my web development roots and have an online portfolio for people to stay updated on my journey

Github Repo

ASCII Image Converter

An application that allows users to submit images and convert them to their ASCII equivalents. Multiprocessing is used to speed up conversion on larger images. The image above is an egg I tested this on.

Github Repo