I have been spending more time learning machine learning, both through personal projects and through my master's program at Georgia Tech. I am enrolled in the OMSCS program, and a lot of the classes I have taken have pushed me to understand data, modeling, and evaluation in a deeper way. Studying ML in grad school has given me a better foundation, and building my own tools has helped me practice those ideas in real ways.
I enjoy learning by building. When I have a real project idea, it keeps me motivated to understand the concepts behind it. Machine learning fits perfectly with the type of finance tools I want to build under Kurt and Leo Labs, so combining school work with personal projects has been a great way to learn faster.
Some of the courses in my program have already helped a lot. In Machine Learning, I built things like decision tree learners, random forests, and bagging algorithms. I also worked on Q learning in another course, which introduced me to reinforcement learning and how agents learn from rewards. In Data and Visual Analytics, I worked on data cleaning, feature selection, correlations, and modeling pipelines. These assignments forced me to think carefully about how data flows and how models need to be evaluated.
I also started learning more about NLP, embeddings, and sentiment scoring. Even though I am still improving in those areas, the class work has given me the confidence to start applying what I know to my own ideas. It also showed me where I need to shape up. For example, I want to get better at deep learning, evaluation metrics, and handling large text datasets. These skills will help me build some of the more advanced tools I want to create in the future.
One of the first ML projects I want to build is a sentiment analysis tool for WallStreetBets. Every earnings season, there is a lot of chatter about certain stocks. Some posts are hype, some are fear, and some actually contain useful insights. The idea is to gather posts, score the sentiment, and show how the mood shifts before earnings.
Even if the predictions are not perfect, it would still be valuable to see patterns. Building this tool would help me practice NLP, text embeddings, and classification models. It would also be fun to compare sentiment to actual price movements, especially for earnings plays. This ties in perfectly with my interest in data, trading, and machine learning.
Another idea I have been working on is an insider purchaser tool. When executives or directors buy their own stock, it can sometimes signal confidence. I want to track insider buys, rank them, and show the most interesting ones. The next step would be using ML to try and classify which ones might lead to short term moves, longer term gains, or which ones should be ignored.
The scope of this idea might be large, and I may need to simplify it along the way. That is part of the learning process. Academic projects at Georgia Tech have shown me that most ideas start bigger than they end up. You refine the scope, adjust the problem, and build step by step. The same approach applies here. Even a smaller version of this tool would still be useful.
The best way I learn is by working through real problems. ML is easier to understand when I can connect the concepts to something practical. The assignments in my classes have given me a strong base, but the personal projects help me apply everything. I get to practice what I learn, try new ideas, and see how models perform in real situations.
Even if these ideas change or grow into something different, the learning stays with me. Every project teaches me something about data, modeling, or design. That is what keeps me moving forward. I am excited to keep improving and exploring more advanced ML topics as I build new tools for Kurt and Leo Labs.
As I learn more about ML and continue through my master's program, I will keep documenting my progress here. These projects are still early, but they have already helped me become a stronger developer.