Designing a multi-modal pipeline to detect abusive content across text, images, and audio. Phase 1 includes data preprocessing, modality-specific feature extraction, model training, and evaluation. Phase 2 will integrate a hybrid CNN-LSTM architecture to improve classification performance.
Developed a Python3 connector using the Meta Graph API to collect posts (text, images, and comments) from a personal Facebook Business Page based on a defined topic. Filtered results are stored in a MongoDB database for further analysis.
Built an interactive chatbot using Hugging Face Transformers and Gradio that remembers previous conversation turns, powered by the FLAN-T5 language model.
Developed a Python script that scrapes faculty profiles from a university website, extracting names, email addresses, and personal web pages using BeautifulSoup and requests.
End-to-end project predicting total invoice amounts using regression models, built from raw SQL data and visualized in Power BI.
Graduation project focused on classifying seizure types from EEG data with optimized channel selection using XGBoost.
Mini version of PixOCR built using the RVL-CDIP dataset with full preprocessing, OCR, and post-processing stages.
Interactive dashboard exploring Netflix content trends by genre, country, and year using Power BI.
Performed customer segmentation and sales trend analysis to guide marketing strategies using SQL and Python.
Exploratory data analysis and predictive modeling to identify customer churn patterns and retention strategies.
Forecasted future product demand using time series analysis and ARIMA models on historical sales data.