My Projects

Created a marketing automation platform tailored for small businesses to manage ad campaigns across Facebook and Instagram.Ad-Pilot streamlines content creation, automated scheduling, real-time performance insights, and competitor analysis to enhance marketing efficiency.

Created a multi-featured WhatsApp bot that transcribes voice messages into text using advanced models like OpenAI's Whisper and Alibaba's ZipEnhancer. Ideal for noisy environments or when listening isn't possible, while also offering smart task scheduling and leveraging Acoustic Noise Enhancement for improved transcription accuracy.

Designed and implemented a Retrieval-Augmented Generation (RAG) pipeline for clinical diagnostic reasoning. Utilized semantic chunking, dynamic confidence scoring, and advanced LLM integration (Palmyra-Med-70B-32k via NVIDIA AI Endpoints). Integrated Google Generative AI Embeddings and Clinical ModernBERT for dual retrieval support. Evaluated with Gemini 2.0 Flash (LLM-as-a-Judge) on MIMIC-IV-Ext clinical notes.

Built an interactive web application that solves mathematical problems written with emojis using a fine-tuned DeepSeek-R1 language model with GRPO (Group Relative Policy Optimization). The model specializes in solving creative mathematical equations represented by emojis, with an easy-to-use Streamlit interface.

Created a deep learning-based Roman Urdu poetry generator using GRU neural networks. It generates ghazals in the style of renowned Urdu poets by leveraging datasets scraped from Rekhta and features an interactive Streamlit web interface.

Developed an Urdu sarcasm detection tool that leverages a Gaussian Naive Bayes model trained on Urdu text data. It features a custom preprocessing pipeline for text normalization, stopword filtering, and tokenization to analyze social media content.

Developed a state-of-the-art text summarization system by fine-tuning BERT, GPT-2, and Llama to generate both extractive and abstractive summaries, leveraging advanced large language models in a Jupyter Notebook environment.

Developed an efficient Arabic-to-English Neural Machine Translation model from scratch by harnessing state-of-the-art Transformers. Applied innovative techniques to deliver high-quality language translation.

AI-powered web application that transforms product images into compelling sales pitches with audio narration. Integrates Qwen VL Max for image analysis, Llama 3 70B for pitch generation, and Zyphra Zonos for TTS, all wrapped in a user-friendly Streamlit interface.

Developed a GPU-accelerated machine learning system to detect fraudulent mobile money transactions. The solution includes extensive EDA, advanced preprocessing (including outlier detection with Isolation Forest and encoding strategies), and model comparison between KNN, SVM, Logistic Regression, and Decision Trees. Achieved 97%+ accuracy with optimal model configurations.

This project provides a comprehensive statistical analysis of data from the Pakistan Super League (PSL) cricket tournament, utilizing various statistical methods and graphical representations such as histograms, pie charts, and regression analysis.

Built a system that repairs mathematically incorrect memes using a fine-tuned Microsoft Phi-4 model. The project includes FastAPI-based inference, Jupyter notebooks for training and testing, and Unsloth optimizations for memory-efficient fine-tuning with LoRA.

This project focuses on applying various Exploratory Data Analysis (EDA) and Feature Engineering techniques to diverse datasets, providing a comprehensive guide for understanding data and preparing it for machine learning models.

This project implements a Next Word Prediction model using Long Short-Term Memory (LSTM) neural networks. It predicts the most likely next word in a sequence of words, providing real-time word suggestions to assist writers and content creators in improving productivity and writing flow. The model is trained on Shakespeare's Hamlet text data.

The Fire-Weather-Index (FWI) Prediction project aims to predict the risk of forest fires by calculating the Fire Weather Index (FWI) using meteorological factors such as temperature, humidity, wind speed, and rainfall. The model is deployed on AWS Elastic Beanstalk, and users can interact with the Flask web app for real-time predictions. The model utilizes Ridge Regression for accurate predictions.

This project develops a machine translation model using Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks to translate English sentences into Urdu. The solution focuses on language nuances and aims to address low-resource language translation challenges.

This project uses an Artificial Neural Network (ANN) to predict customer churn for banks, helping them take proactive measures to retain at-risk customers. It features an interactive Streamlit web app for real-time predictions and insights.

This project explores advanced prompt engineering techniques (Zero-shot, One-shot, and Few-shot prompting) and fine-tuning methods across various language models like Gemma, Llama, and DistilBERT. It leverages datasets such as databricks-dolly-15k and applies techniques like LoRA-based fine-tuning for enhanced model performance.

A bidirectional converter that translates between C++ code and pseudocode using transformer-based neural networks. The project implements both code-to-pseudocode and pseudocode-to-code translation capabilities using PyTorch, along with an interactive web interface built with Streamlit.