multi agent radiology assistant

Client: qatar hospitals

Multi-agentic ai assistant for hotel bookings and recommendation

Client: St regis doha

Developed an AI-powered hospitality assistant for Amara Hospitality with a TypeScript frontend and OpenAI-based agents. The assistant understands guest queries, recommends rooms based on preferences and available data, dynamically displays relevant rooms and amenities, and guides users through the website without manual navigation.

Almanai-group-heavy equipements sales forecasting and recommendations

Client: almanai-group

Developed an interactive sales dashboard for Al Manai Heavy Equipment to monitor KPIs, inventory, brands, sales performance, and AI-driven business insights.

almanai group cars sales forecasting

Client: almanai-group

AI-powered automotive sales platform with interactive dashboards, 12-month sales forecasting, customer analytics, churn prediction, inventory optimization, and AI-driven business recommendations.

Qatar museums footfall and sales forecasting

Client: Qatar museums

AI-powered visitor and revenue forecasting dashboard built with Streamlit. The system combines Prophet and Random Forest models to predict museum visitors and revenue for the next 30–180 days, while analyzing seasonality, weekends, and temperature. It provides interactive visualizations, peak-period detection, revenue breakdowns, strategic recommendations, and Excel report generation.

Amara Hospitality booking agent

Client: Katara-hospitality

Developed Amara, a conversational AI chatbot that understands user queries, provides relevant answers, and delivers personalized recommendations based on user preferences and conversation context. Built with a React frontend for a responsive, real-time chat interface and powered by the OpenAI API for natural language understanding and response generation, Amara interprets user intent, maintains conversational context, and tailors its suggestions to each user's needs — making it well-suited for use cases like personalized recommendations, Q&A support, and interactive guidance.

Capital technology solutions chatbot

Client: Capital technology solutions

Developed an AI chatbot using RAG to answer questions from company documents. Implemented OpenAI for LLM-based responses, FAISS for vector search and document retrieval, and FastAPI for backend API development.

AI trip planner multi-agent

Client: freelancing project

AI-powered trip planner that uses multi-agent workflows to create personalized itineraries with real-time travel, weather, currency, and destination information.

AI multi agent cv recruitement system

Client: freelancing project

Developed a multi-agent recruitment system using LangGraph and LangChain to automatically analyze CVs, evaluate candidate skills, experience, and cultural fit, calculate candidate scores, and generate hiring recommendations through an interactive Streamlit interface.

Cats Breed prediction

Client: personal project

Developed a deep learning application for real-time cat breed classification, allowing users to upload images and identify cat breeds through a user-friendly interface.

Cars number plate detection and recognition

Client: Horizop energy

AI-powered license plate recognition and EV platform for detecting and recognizing vehicle plates, retrieving vehicle information, and managing EV charging stations and sessions.

diabet disease prediction

Client: Freelancing project

Developed a machine learning classification model to predict diabetes risk in patients using the PIMA Indians Diabetes dataset. The project covers the full ML pipeline — from exploratory data analysis and preprocessing to model training, evaluation, and deployment-ready serialization. Built a Support Vector Machine (SVM) classifier with a linear kernel that analyzes 8 key health indicators (glucose level, BMI, blood pressure, age, etc.) to classify patients as diabetic or non-diabetic, achieving strong accuracy on unseen test data. The trained model was serialized with Pickle to enable reuse in a predictive system without retraining.

Multi-agent-ai fertility specialised assistant

Client: Tunsian and north african clinics

Built an AI-powered medical fertility assistant using a GraphRAG architecture to provide context-aware, knowledge-grounded responses. Designed and implemented LangGraph multi-agent workflows for specialized medical reasoning, entity extraction, retrieval, and response generation. Developed medical entity extraction pipelines using SciSpaCy and spaCy, and integrated LLMs for intelligent conversational assistance. Implemented the backend as scalable microservices and established automated CI/CD pipelines with Azure DevOps.

Toxic comment detection application

Client: Smart conseil company project

Built a multi-label NLP classification system to detect toxic behavior in online comments, using the Jigsaw Toxic Comment Classification dataset. The project identifies six categories of harmful content — toxic, severe toxic, obscene, threat, insult, and identity hate — enabling automated content moderation for platforms dealing with user-generated text. The pipeline covers full text preprocessing (cleaning special characters, punctuation, non-ASCII text), exploratory data analysis with visualizations (class distribution plots, word clouds per toxicity category), and class balancing to address heavy label imbalance. Multiple machine learning models — Logistic Regression, K-Nearest Neighbors, Naive Bayes (Multinomial & Bernoulli), Linear SVM, and Random Forest — were trained and benchmarked using TF-IDF vectorization, with F1-score used to select the best-performing model per category. The final Random Forest model was deployed for real-time probability-based toxicity scoring on new comments and serialized for reuse.