I have developed a Chrome extension for language learning that retrieves context from subtitles or text in articles and provides the meaning of the word in question within that context. The extension utilizes the Mistral API to generate responses in various languages. Additionally, it incorporates an Anki flashcard feature.

AntiForget.app is a revision web app that utilizes a custom spaced repetition algorithm, and ai-generated-quiz to test your knowledge of the uploaded topics, and based on the feedback, the topic will be positioned in the revision queue.

Built an NLP pipeline using a Mixture-of-Adapters (MoA) on RoBERTa to detect complex sarcasm across distinct English dialects from BESSTIE dataset
Fine-tuned Mistral-7B using Parameter-Efficient Fine-Tuning (LoRA) and a novel Tensor-of-Cues (VAToC) strategy outperforming baseline on BESSTIE benchmark

Curated a comprehensive Natural Language Inference (NLI) dataset containing 42,889 instances specifically for the biomedical domain.

Accurate point predictions are not enough for financial models; reliable probability estimates are equally critical. To address this, we applied a perturbation-based consistency calibration method (C3) which improved the Expected Calibration Error (ECE), resulting in highly stable and reliable confidence estimates.

A two-step framework bridging Vision-Language Models (Video-LLaVA) and segment localization to extract precise textual answers from unstructured, egocentric videos.

A novel hybrid Machine Learning framework combining Convolutional Neural Networks (CNNs) and Random Forests for hyperspectral image oil spill classification.

This project involves the development of a multi-output pipeline for predicting the coordinates (x, y) of a particle passing through a resistive silicon detector (RSD). The pipeline uses the XGBRegressor to analyze signals generated by 12 pads in the detector.