Visualization of financial metrics extracted from the SEC EDGAR 10-K filings database across a 20-year timespan for multiple tickers. Extraction is performed using Google Gemini 1.5 Flash (along with a bit of prompt engineering) and regex expressions to flexibly support various filing formats. Finished in ~3 days.

LLM + 10-K

Revived and maintained the LC3Tools project, an educational software suite to write, simulate, and debug assembly for the Little Computer 3. Worked on this while I was a CS 2110 TA, used by over 1000 students per semester. Completely revamped its testing suite, resolved many tokenizer/assembler nuances, and made countless QoL UI improvements.

View the gt-cs2110 fork of LC3Tools here.

LC3Tools

Final project for Georgia Tech's Applications to Linguistics (LING 3100) course. Used a RoBERTa model fine-tuned to classify 28 different emotions to evaluate trends of sentiments of U.S. Presidential Inaugural Addresses over the course of history. Used BeautifulSoup 4 to scrape transcripts of these Addresses from the American Presidency Project website, along with Hugging Face Transformers and NLTK for the actual sentiment analysis.

POTUS Sentiment Analysis

Semester-long project on utilizing various machine and deep learning techniques, most notably PCA, CNNs, and transfer learning, to classify "snacks" in an image frame as accurately as possible. Experimented with ResNet, Vision Transformer, image reduction/compression methods, and many more.

Snacks Image Classification

A deep learning project to automate the labeling of caribou instances in Alaskan camera trap data using Foreground Segmentation. Completed under the supervision of Dr. Michael Mandel at Brooklyn College CUNY. Has won the Terra NYC STEM First Award and the Milton Fisher Scholarship for Innovation and Creativity.

Paper_21_6_SO.pdf

View the project's Github, Paper, and Video.

Projects

LLM + 10-K

LC3Tools

POTUS Sentiment Analysis

Snacks Image Classification

Caribou Image Annotator & Detector