This project, ranking 5th solo among 236 teams, aims to predict Europe's next-hour renewable energy surplus. It utilizes XGBoost, LightGBM, and LSTM models, trained on 2022 data and tested on early 2023 data, providing key insights into renewable energy forecasting trends.
An advanced OCR pipeline designed to process book pages, featuring image dewarping, preprocessing, and OCR text extraction using tools like page-dewarp, OpenCV, and Tesseract, all demonstrated in a Streamlit app.
Implementation of Lamsiyah et al. (2021) for unsupervised multi-document extractive summarization, employing transformer and compression-based embeddings, supported by a three-dimensional scoring system for document/sentence relevance.
Property Graphs (utilizing Neo4j), Distributed Graphs (leveraging Pregel), Knowledge Graphs (employing GraphDB), and a culminating project that integrates these graph databases for machine learning applications.
An extension of Andrej Karpathy's Micrograd to include support for Recurrent Neural Networks (RNNs) in addition to standard neural networks. This project provides a minimalistic yet powerful way to understand and experiment with RNNs.
Custom fine-tuned BERT model using HuggingFace for extractive summarization, along with a utility to convert abstractive to extractive data, aimed at enhancing model training and interpretability.
An automated parser for transcribing Kindle notes and highlights directly into a Notion page, simplifying the process of consolidating and revisiting key takeaways from readings.
A time-series analysis to predict the number of flu cases in the USA. The framework is based on ARIMA models. It consists of pattern recognition, model estimation, statistical validation, predictive analysis, and outlier management.
An end-to-end big data pipeline with Real Estate data. It features raw data ingestion into Apache Hadoop, storage in MongoDB, and data transformation, machine learning, and real-time data streaming with Apache Spark. Final KPIs are visualized in Tableau.
This project develops a data management and analysis pipeline to cluster countries based on multiple indicators such as economical, geo-political, and governmental data. SQL is used for the data warehouse, Python for the data analysis.