My Projects
My Projects
check my github
MiruAI: AI-Powered Virtual Try-On & E-Commerce Fitting Platform
(ML PROJECT)
MiruAI: AI-Powered Virtual Try-On & E-Commerce Fitting Platform
(ML PROJECT)
An AI-powered virtual try-on and e-commerce platform that allows users to visualize garments on their own body. The core virtual try-on model was trained on Google Cloud Platform (GCP) and deployed as a production inference service on a Linux VPS, making the AI capability accessible through the web application.
The platform integrates the trained generative model into an end-to-end try-on workflow, where users can provide their body profile or photo, select garments, and generate AI-powered previews of the clothing on their body. The system also supports combining multiple garments into complete outfits through a dedicated fitting-room experience.
The application was built with Next.js and PostgreSQL, with the AI inference service integrated into the platform through an API-based architecture. The deployment on a Linux VPS provides a production environment for serving the trained model and handling try-on requests.
Technologies: Python, Generative AI, Computer Vision, Deep Learning, GCP, FastAPI, Next.js, PostgreSQL, Linux, Docker.
An AI-powered virtual try-on and e-commerce platform that allows users to visualize garments on their own body. The core virtual try-on model was trained on Google Cloud Platform (GCP) and deployed as a production inference service on a Linux VPS, making the AI capability accessible through the web application.
The platform integrates the trained generative model into an end-to-end try-on workflow, where users can provide their body profile or photo, select garments, and generate AI-powered previews of the clothing on their body. The system also supports combining multiple garments into complete outfits through a dedicated fitting-room experience.
The application was built with Next.js and PostgreSQL, with the AI inference service integrated into the platform through an API-based architecture. The deployment on a Linux VPS provides a production environment for serving the trained model and handling try-on requests.
Technologies: Python, Generative AI, Computer Vision, Deep Learning, GCP, FastAPI, Next.js, PostgreSQL, Linux, Docker.
RetinAI: AI-Powered Diabetic Retinopathy Screening
(ML PROJECT)
RetinAI: AI-Powered Diabetic Retinopathy Screening
(ML PROJECT)
An end-to-end clinical AI platform that combines deep learning-based retinal image classification with a multi-agent AI pipeline for explainable screening and automated clinical reporting.
The system uses EfficientNet-B4 / Vision Transformer models to classify diabetic retinopathy across five severity stages and generates Grad-CAM heatmaps to provide visual explanations of model predictions. A LangGraph-based multi-agent pipeline coordinates specialized agents for clinical interpretation, Grad-CAM analysis, PubMed-based RAG, and automated medical report generation.
The platform exposes the analysis through a FastAPI backend and a Next.js frontend, with PostgreSQL for patient records and ReportLab for generating structured PDF reports.
Technologies: PyTorch · EfficientNet-B4 · Vision Transformer · Grad-CAM · LangGraph · LangChain · RAG · PubMed API · Llama 3.3 · Gemini 2.0 Flash · FastAPI · Next.js · PostgreSQL · Docker · ReportLab
An end-to-end clinical AI platform that combines deep learning-based retinal image classification with a multi-agent AI pipeline for explainable screening and automated clinical reporting.
The system uses EfficientNet-B4 / Vision Transformer models to classify diabetic retinopathy across five severity stages and generates Grad-CAM heatmaps to provide visual explanations of model predictions. A LangGraph-based multi-agent pipeline coordinates specialized agents for clinical interpretation, Grad-CAM analysis, PubMed-based RAG, and automated medical report generation.
The platform exposes the analysis through a FastAPI backend and a Next.js frontend, with PostgreSQL for patient records and ReportLab for generating structured PDF reports.
Technologies: PyTorch · EfficientNet-B4 · Vision Transformer · Grad-CAM · LangGraph · LangChain · RAG · PubMed API · Llama 3.3 · Gemini 2.0 Flash · FastAPI · Next.js · PostgreSQL · Docker · ReportLab
Customer Churn & Lifetime Value Analytics
(DATA ANALYTICS PROJECT)
Customer Churn & Lifetime Value Analytics
(DATA ANALYTICS PROJECT)
End-to-end customer analytics project analyzing 10,000 retail banking customers to identify churn drivers, quantify revenue at risk, predict customer churn, and develop targeted retention strategies. The project combines Python-based data analysis, statistical hypothesis testing, survival analysis, machine learning, customer risk scoring, and Lifetime Value (LTV) modeling with interactive business intelligence dashboards built in Power BI. The analysis identifies high-risk customer segments, evaluates churn patterns across demographics, geography, product holdings and engagement, estimates potential revenue loss, and supports data-driven retention decisions through customer-level risk and value segmentation. Technologies: Python, Pandas, SciPy, Scikit-learn, Lifelines, Power BI, DAX, Power Query, Docker.
End-to-end customer analytics project analyzing 10,000 retail banking customers to identify churn drivers, quantify revenue at risk, predict customer churn, and develop targeted retention strategies. The project combines Python-based data analysis, statistical hypothesis testing, survival analysis, machine learning, customer risk scoring, and Lifetime Value (LTV) modeling with interactive business intelligence dashboards built in Power BI. The analysis identifies high-risk customer segments, evaluates churn patterns across demographics, geography, product holdings and engagement, estimates potential revenue loss, and supports data-driven retention decisions through customer-level risk and value segmentation. Technologies: Python, Pandas, SciPy, Scikit-learn, Lifelines, Power BI, DAX, Power Query, Docker.
Road Traffic Injury Accidents Analysis
(DATA ANALYTICS PROJECT)
Road Traffic Injury Accidents Analysis
(DATA ANALYTICS PROJECT)
Exploratory and statistical analysis of 309,756 road accidents in France (2019–2024) to identify when, where, and why accidents become more severe. Built data visualizations and statistical tests to uncover patterns related to road conditions, time, location, vehicle type, weather, and safety equipment.
Exploratory and statistical analysis of 309,756 road accidents in France (2019–2024) to identify when, where, and why accidents become more severe. Built data visualizations and statistical tests to uncover patterns related to road conditions, time, location, vehicle type, weather, and safety equipment.
Real-Time Fraud Detection Lakehouse
(DATA ENGINEERING PROJECT)
Real-Time Fraud Detection Lakehouse
(DATA ENGINEERING PROJECT)
An end-to-end data engineering project designed to process and monitor credit card transactions as a real-time streaming workload. The pipeline replays 284,807 real credit card transactions through Redpanda, validates and ingests events into a Delta Lake on MinIO object storage, and processes them through a Bronze → Silver → Gold Medallion architecture using DuckDB SQL.
The Silver layer performs data cleansing, deduplication, enrichment, transaction-velocity analysis, anomaly detection, and fraud risk scoring. Gold models produce business-ready metrics such as hourly fraud rates, merchant risk summaries, and high-risk account rankings. A Streamlit dashboard provides real-time monitoring of pipeline throughput, fraud alerts, precision/recall, and risk analytics.
The entire platform is containerized with Docker Compose and includes automated testing and data-quality validation.
Technologies: Python · Redpanda · Kafka API · Delta Lake · MinIO · DuckDB · SQL · Docker · Docker Compose · Streamlit · Pytest
An end-to-end data engineering project designed to process and monitor credit card transactions as a real-time streaming workload. The pipeline replays 284,807 real credit card transactions through Redpanda, validates and ingests events into a Delta Lake on MinIO object storage, and processes them through a Bronze → Silver → Gold Medallion architecture using DuckDB SQL.
The Silver layer performs data cleansing, deduplication, enrichment, transaction-velocity analysis, anomaly detection, and fraud risk scoring. Gold models produce business-ready metrics such as hourly fraud rates, merchant risk summaries, and high-risk account rankings. A Streamlit dashboard provides real-time monitoring of pipeline throughput, fraud alerts, precision/recall, and risk analytics.
The entire platform is containerized with Docker Compose and includes automated testing and data-quality validation.
Technologies: Python · Redpanda · Kafka API · Delta Lake · MinIO · DuckDB · SQL · Docker · Docker Compose · Streamlit · Pytest
Olist E-Commerce Data Warehouse: Airflow, dbt & PySpark
(DATA ENGINEERING PROJECT)
Olist E-Commerce Data Warehouse: Airflow, dbt & PySpark
(DATA ENGINEERING PROJECT)
Built an Airflow-orchestrated ELT pipeline ingesting the Olist e-commerce dataset (9 relational tables, 100k orders) with
PySpark into a PostgreSQL data warehouse, fully containerized with Docker Compose.
Modeled a star schema (fact and dimension tables) with dbt, including incremental models, automated data tests, and docu-
mentation; added CI checks with GitHub Actions.
Built an Airflow-orchestrated ELT pipeline ingesting the Olist e-commerce dataset (9 relational tables, 100k orders) with
PySpark into a PostgreSQL data warehouse, fully containerized with Docker Compose.
Modeled a star schema (fact and dimension tables) with dbt, including incremental models, automated data tests, and docu-
mentation; added CI checks with GitHub Actions.
Library Management System
(FULL STACK PROJECT)
Library Management System
(FULL STACK PROJECT)
Full-stack library management application built with React, Flask, and Oracle Database, featuring role-based access, authentication, book management, and complete borrowing workflows with PL/SQL-backed business logic.
Technologies : React · Vite · JavaScript · Python · Flask · Oracle Database · PL/SQL · REST API · SQL
Full-stack library management application built with React, Flask, and Oracle Database, featuring role-based access, authentication, book management, and complete borrowing workflows with PL/SQL-backed business logic.
Technologies : React · Vite · JavaScript · Python · Flask · Oracle Database · PL/SQL · REST API · SQL
Card Scanner & OCR Contact Extraction
(COMPUTER VISION / OCR)
Card Scanner & OCR Contact Extraction
(COMPUTER VISION / OCR)
An automated document-processing application designed to digitize batches of business cards. Users can upload multiple cards and the system uses EasyOCR and OpenCV to detect and extract contact information including names, companies, phone numbers, emails, and addresses.
The extracted information is cleaned and structured using Pandas and OpenPyXL, then exported directly into a formatted Excel spreadsheet. The application supports both Arabic and English text, validates uploaded files, and provides a web interface for batch processing.
The system is built with a Next.js/TypeScript frontend and a FastAPI Python backend, with Docker support for deployment.
Technologies: Python · FastAPI · EasyOCR · OpenCV · Pandas · OpenPyXL · Next.js · TypeScript · Tailwind CSS · Docker
An automated document-processing application designed to digitize batches of business cards. Users can upload multiple cards and the system uses EasyOCR and OpenCV to detect and extract contact information including names, companies, phone numbers, emails, and addresses.
The extracted information is cleaned and structured using Pandas and OpenPyXL, then exported directly into a formatted Excel spreadsheet. The application supports both Arabic and English text, validates uploaded files, and provides a web interface for batch processing.
The system is built with a Next.js/TypeScript frontend and a FastAPI Python backend, with Docker support for deployment.
Technologies: Python · FastAPI · EasyOCR · OpenCV · Pandas · OpenPyXL · Next.js · TypeScript · Tailwind CSS · Docker
Responder: AI-Powered Google Review Automation
(AI ENGINEERING / GENERATIVE AI)
Responder: AI-Powered Google Review Automation
(AI ENGINEERING / GENERATIVE AI)
A multi-business AI automation platform that manages Google Business Profile reviews from ingestion to response publication. Reviews are automatically classified by rating and sensitivity, then routed through different workflows: positive reviews receive automatically generated responses, neutral reviews require human approval, while negative or sensitive reviews are escalated directly to a human.
The platform uses OpenAI or Google Gemini to generate contextual responses while enforcing business-specific context as the model's only source of factual information, reducing the risk of hallucinated claims. Each business has isolated credentials, configuration, brand voice, and review data.
The system integrates Google OAuth and Business Profile APIs, automated cron-based ingestion and publishing, a Prisma/SQLite database, encrypted credentials using AES-256-GCM, and a Next.js dashboard for approvals, escalations, settings, and monitoring.
Technologies: Next.js 15 · TypeScript · React · Tailwind CSS · Prisma · SQLite · Google APIs · OAuth 2.0 · OpenAI · Gemini · Node.js · node-cron · AES-256-GCM
A multi-business AI automation platform that manages Google Business Profile reviews from ingestion to response publication. Reviews are automatically classified by rating and sensitivity, then routed through different workflows: positive reviews receive automatically generated responses, neutral reviews require human approval, while negative or sensitive reviews are escalated directly to a human.
The platform uses OpenAI or Google Gemini to generate contextual responses while enforcing business-specific context as the model's only source of factual information, reducing the risk of hallucinated claims. Each business has isolated credentials, configuration, brand voice, and review data.
The system integrates Google OAuth and Business Profile APIs, automated cron-based ingestion and publishing, a Prisma/SQLite database, encrypted credentials using AES-256-GCM, and a Next.js dashboard for approvals, escalations, settings, and monitoring.
Technologies: Next.js 15 · TypeScript · React · Tailwind CSS · Prisma · SQLite · Google APIs · OAuth 2.0 · OpenAI · Gemini · Node.js · node-cron · AES-256-GCM