Analyzed cross-lingual knowledge representation in BLOOM-1.7B by computing cosine similarity across English, French, and Portuguese, achieving high average similarity (0.91-0.94) despite Portuguese having only 5.2% share in training data.

• Created a trilingual lexicon dataset with Gemini API, available on Hugging Face.

   https://huggingface.co/datasets/MLap/English-French-Portuguese-Lexicon

• Conducted zero-shot cross-lingual transfer experiments by fine-tuning BLOOM on (High Resource   

  Language) English sentiment classification task and testing on Hindi (Low Resource Language) to 

  evaluate multilingual generalization capabilities.

• Created a SentiHin-2500 (Hindi Sentimental Analysis dataset), available on Hugging Face.

   https://huggingface.co/datasets/MLap/SentiHin-2500
    Visit GitHub repo

• Main feature: Haze removal from image/ video.

• Demo at 🤗:  huggingface.co/spaces/MLap/deFogify

• This project implements a single-image haze removal technique using the Dark Channel Prior, as described in the research paper: ``Single Image Haze Removal Using Dark Channel Prior`
Visit GitHub repo

• 30+ ⭐ on GitHub

• Extracts links, text, and more from custom screenshots.

• Advantages: Requires "no internet", free to use.

•  Integrated with Google Gemini through its API for query on extracted text. [This Requires Internet]

• Some features developed in '36 hours' at (IIT Dhanbad) Hack'24.

• Utilizes LSTM-based Tesseract OCR library for text recognition from images.
  Visit GitHub repo