Projects

Project Dashboard

Last updated: July 28, 2025

• Projects by Year

2025

Models on Hugging Face

• LoRA Model (Phi‑4‑based English Chatbot) – Lightweight chatbot fine‑tuned with LoRA and Unsloth on FineTome‑100k, 4‑bit quantized .
• LLaMA 3.2‑3B Chat Model – Instruction‑tuned 3B conversational model supporting 2048‑token context, LoRA‑trained .
• Vision‑Language Medical Captioning Model – Fine‑tuned on radiology images via QLoRA and LoRA adapters for image‑to‑text tasks .
• Kashur GPT (Kashmiri Model) – Transformer-based decoder trained for Kashmiri text generation with demo available via Space .
• English SpeechT5 TTS – Fine‑tuned SpeechT5 on LJSpeech + custom data, optimized for technical English pronunciation .
• Turkish SpeechT5 TTS – Turkish TTS fine‑tuned on native dataset with MOS evaluation, implemented as part of internship project .
• Kashmiri LLM BERT‑Base – Experimental Kashmiri BERT‑base model trained from scratch (~45 K corpus), mainly for research .

---

Datasets on Hugging Face

• vision_language_pairs_data – Multimodal image‑text dataset for vision-language modeling
• environmental_sound_data_hnm – Environmental sound recordings dataset for audio classification
• human_genome_variants_hnm – Annotated genomic variants for bioinformatics research
• BITCOIN_TRANSACTION_DATASET_hnm – Blockchain transaction records for financial analysis
• kashmiri_multilingual_dictionary_dataset – Parallel dictionary dataset for translation and NLP tasks
• fairness_benchmark_data_HNM – Bias/fairness benchmarking dataset for ethical ML evaluation
• 17k‑high‑quality‑stack‑overflow‑QA‑pairs – Cleaned QA dataset for programming Q&A use cases
• Kashmiri_Text_Corpus_Cleaned_2025_HNM – Clean Kashmiri corpus curated in 2025 for NLP research
• kashmiri_sharegpt – ShareGPT-format conversational dataset for Kashmiri chatbot fine-tuning
• 200_Kashmiri_food_dishes – Cultural dataset of Kashmiri dish names/details for NLP tasks
• Kashmiri Text Corpus Dataset – Core Kashmiri corpus for academic NLP research
• Urdu Text Corpus (Sentence wise) Dataset – Sentence-level Urdu corpus for NLP study
• Persian Text Corpus Dataset – Persian language corpus for academic NLP tasks
• 40K Kashmiri text recognition dataset – OCR dataset with 40K image-text pairs for text recognition
• 30K Kashmiri text recognition dataset – OCR dataset (~30K samples) for Kashmiri script detection
• Urdu News Text Corpus Dataset – Urdu news articles corpus for research and NLP
• English TTS Dataset – Spoken English dataset curated for TTS training

---

Spaces by HNM (Omarrran)

• PyTorch‑8‑Bit‑Weight‑Quantizer – Web app to quantize neural network weights to 8-bit precision .
• Video_Rackground_Removal – In-browser real-time background removal using Transformers.js and MODNet .
• DSPy Summarization HNM – PDF uploader summarizer Space that chunks and summarizes documents .
• II Eleven Labs Speech – Speech synthesis demo using Eleven Labs API key input functionality .

---

Note: For the most up‑to‑date list of models, datasets, and Spaces, please visit the Omarrran profile on Hugging Face.

2024 – Best Projects (TTS, Fine-Tuning, BERT, Deep Learning)

• Fine-Tuning Microsoft SpeechT5 TTS for Turkish:

  This project focuses on customizing Microsoft’s SpeechT5 TTS model for Turkish language synthesis, showcasing advanced multilingual speech applications.

  View Project: Turkish TTS Fine-Tuning

• Training BERT for Sentiment Analysis:

  Builds on the BERT transformer architecture to classify sentiment (positive, negative, neutral) with high accuracy.

  View Project: BERT IMDB Reviews

• Deep Learning-based Inpainting & Object Removal:

  An advanced approach that uses CNNs for removing objects or filling in missing regions in images seamlessly.

  View Project: Inpainting Using Flux

---

2023

• Music Genre Classifier via CNN:

  Classifies audio clips into different genres using convolutional neural networks and MFCC feature extraction.

  View Project: Music Genre Classifier

• Human Activity Recognition:

  Detects user activities (walking, standing, etc.) using smartphone sensor data and machine learning models.

  View Project: Human Activity Recognition

---

2022

• Credit Card Fraud Detection:

  Utilizes machine learning techniques to identify fraudulent transactions within real-world financial datasets.

  View Project: Credit Card Fraud Detection

• Detecting Fake News:

  Builds a text classification model to distinguish real news articles from fake ones.

  View Project: Fake News Detection

• Personal Movie Recommendation with LGBM Ranker:

  Provides tailored movie suggestions based on user preferences using LightGBM for ranking.

  View Project: Movie Recommendator

---

2021

• Spam Email Detection with Deep Learning:

  Utilizes neural networks and NLP preprocessing to classify emails as spam or ham.

  View Project: Spam Detection

• Face Recognition using LBPH:

  Implements OpenCV’s LBPH for recognizing faces in real-time camera feeds.

  View Project: Face Recognition

• Handwritten Digit Classifier (MNIST):

  Employs a simple neural network for digit classification on the MNIST dataset.

  View Project: Digit Classifier

Some Fun and Learning Projects

• AI LiveKit Voice Assistant

  A voice assistant using LiveKit for real-time interaction with Deepgram (speech-to-text), OpenAI (LLM + TTS), and Silero (VAD).

  View Project: AI LiveKit Voice Assistant

• Object Eraser

  Automatically remove objects from images by naming them or drawing bounding boxes, seamlessly handling shadows/reflections.

  View Project: Object Eraser

• Math Solver with Gradio

  Web app that uses OCR (Qwen2-VL) & NLP (Qwen2-Math) to solve math queries from images or text, all via a Gradio interface.

  View Project: Math Solver with Gradio

• Python Discord Bot

  A simple Discord bot built with discord.py to demonstrate basic command handling and user interaction.

  View Project: Python Discord Bot

• Brain Tumor Detection

  Applies image processing and ML models to MRI scans for brain tumor segmentation and classification.

  View Project: Brain Tumor Detection

• OCR with Multiple Languages

  Extracts text (in various languages) from images and offers text search within extracted content.

  View Project: Multi-Language OCR

• Healthcare Chatbot

  Machine learning–based symptom diagnosis system using decision tree and SVM, offering precaution recommendations.

  View Project: Healthcare Chatbot

• Speech Emotion Recognition

  Analyzes audio features to classify emotional states (e.g., happy, sad, angry) using deep learning.

  View Project: Speech Emotion Recognition

• Stock Price Prediction

  Predicts next-day stock behavior (up/down) from historical market data using various ML models.

  View Project: Stock Price Predictor

• Hate Speech Detection

  Classifies tweets into hate speech, offensive language, or neither using text preprocessing and Keras-based models.

  View Project: Hate Speech Detection

---