VENKAT GELLA

I am machine learning enthusiast with a background in Computational Neuroscience and Computer Science. I have research experience working on highly dimensional, biological datasets and algorithm design. My personal projects have more recently included exploring custom trained GPT networks for specific tasks.

rXivgpt (Python) I scraped all of 2023's Biorxiv Neuroscience articles to train a custom, local gpt network to answer questions on neuroscience and generate novel neurosxience ideas. I used huggingface and Pytorch libraries to develop a method for training from gpt2 weights to produce high-quality, custom gpt frameworks.

Cognitive Robotics Project (Python) Using the Cozmo robot, I developed a python project to use cozmo to manipulate magnetic cubes (made from cardboard cutouts) into any 2d structure on the floor in a physical environment. I used chatgpt3.5 in the backend for the user experience so that the user could ask for any 2d object, even using colloqial terms such as "draw a W", and chatgpt would organize that information into a python object structure representation which was further parsed to build a code representation to be interpreted and built by the robot in the physical environment.

Stock Market Analysis Machine Learning Models (Python). I combined web-scraping data with historical earnings reports to train neural network models and preliminary Astronet implementation to predict surprises in quarterly stock earnings. In this project, I combined chatgpt based analysis of news article titles with a GRU time sequence machine learning prediction of prices to ultimately predict whether an earnings report would exceed expectations.

Additional projects I have worked on a wealth of other projects in languages including Matlab, Java, SML, C, and other languages on topics ranging from compiler development to algorithm design to statistical analysis to software development.