Headshot

I'm Tommy Adams.
A software engineer
based in the SF Bay Area.

M.S. Software Development (2025)

Python | SQL | Pytorch | Pandas | Git

Java | JavaScript | React | Node | Excel

About Me

I am a software engineer with a strong foundation in data analysis, machine learning, and full-stack development. I hold B.A.s in Statistics and Economics from UC Berkeley and am pursuing an M.S. in Software Development from Boston University (expected April 2025). I am currently a Machine Learning Research Assistant at Boston University, where I am exploring the trade-offs between model performance and compute efficiency for neural network ensembles. Previously, I worked at New Relic, where I built ETL data pipelines and anomaly detection models to optimize cloud costs.

Outside of research, my machine learning projects include building a generative transformer language model, fine-tuning LLMs using QLoRA, and implementing RAG with vector search. I have also designed and deployed multiple web applications, including a Twitter-like social platform.

On weekends, I enjoy hiking with my dog, discovering new board games, drafting MTG cards, playing ultimate frisbee and pickleball, and hosting poker nights with friends.

My Work

Transformer Language Model

Constructed a generative pre-trained transformer (GPT) language model in PyTorch from scratch. Engineered a multi-transformer architecture with multi-head attention to optimize performance. Trained the model to generate new song lyrics based on a database of Taylor Swift songs.

RAG Vector Search

Enhanced Chat GPT's domain-specific knowledge of 20,000 Leagues Under the Sea by implementing retrieval-augmented generation (RAG) with context derived from a NoSQL vector database. Crafted a unique conversational style inspired by Captain Jack Sparrow via LLM prompt engineering.

Fine Tuning TinyLlama

Utilized Hugging Face transformers and Peft packages to fine-tune a 1.1B parameter large language model to generate song lyrics by implementing QLoRA and double quantization techniques.

Stock Market Groupings

Identified new groupings for S&P 500 stocks based on stock price movement. Utilized a K-means algorithm in PyTorch to create the groupings. Prepared data with Pandas and graphed results with Matplotlib. Introduced novel groupings that outperformed industry standards, revealing insightful industry correlations.

Generative Adversarial Network

Built a generative adversarial neural network (GAN) in PyTorch. Trained the network to produce unique images of apples and detect fake images. Achieved a broad variety in generated apple images with varying positions, colors, and lighting conditions.

Messaging Web App

Created a Twitter-like web app with public posts, news feed, likes, account creation, and user verification. Generated thousands of posts and profile images with Chat GPT and DALL-E to test app functionality.

Call Logger

Built and deployed a web app for summarizing meeting notes with React, Node, Express, and Chat GPT API. Call Logger allows users to upload a series of calls or meeting notes and query the documents. The query is answered with a list of facts that the user can individually approve or reject.