High School (Columbia Summer Program)

I built a Chrome extension to filter political tweets from Twitter after training a model to detect political content during a data science summer course at Columbia.

This was the summer before my senior year of high school. For the final project in the course, my team worked on classifying whether tweets were political or not. We collected datasets of political tweets (election content, political discussions) and non-political tweets, cleaned the data, and trained several models.

We tried a logistic regression with TfidfVectorizer first, then built a neural network with an embedding layer, bidirectional LSTM, ReLU activation, and a dense output layer. We also experimented with pre-trained GloVe embeddings. The model performed well on the test data.

I built a Chrome extension that would use the model to filter tweets. The extension would scan tweets as they loaded on the page and hide anything the model flagged as political.

Later, I added functionality to connect to local LLMs so I could filter other content too. For example, I could have it remove Real Madrid tweets or any other topic I didn't want to see. Instead of training specific models for each thing, I just connected to a local LLM and had it determine if a tweet contained whatever I wanted filtered out.

It was a practical way to take a class project and turn it into something I actually used.