Sari Itani

Highly motivated and results-oriented Computer Engineer, specialized in Artificial Intelligence and passionate about backend development, constantly seeking a challenging position to leverage my technical expertise and contribute to intelligent and efficient computer systems.

Founder and President

2023 - Present · ARC & IEEE-CIS

Co-founder

2023 - Present · LOOP

IdeaLab AI Hackathon 2024 [Team EV-SOL]

June 2, 2024 · IdeaLab

Idea: Electric Vehicle industry suffers from proper planning of power distribution straight from power source to the charging stations. EV-SOL suggests a strategy that helps cut the energy loss, and thus help save energy which means cutting costs and helping the environment, by providing the EV industry with 2 data driven AI models to track the ideal positions of EV charging stations, and for predicting the hourly power demand to be routed from the generators to the stations across the smart power grid."
View Pitch: Click here

LAU IEEE-Robotics D.A.R.E [Team Scotlia]

April 27, 2024 · Lebanese American University

Idea: Developing a cheap yet efficient wildfire surveillance system to quickly scan the area and generate a priority zone that would instruct fire fighters to most optimally stop the fire, based on aerial images and environmental data and critical conditions such as presence of casualties and human residence, hence saving time and lives.

LAU Case Competiton

June 9, 2023 · Lebanese American University

Idea: ''Revolutionizing Lebanese Tourism: Embrace the Hunt Campaign'' Spearheaded the development and execution of an AR-enabled scavenger hunt campaign aimed at promoting Lebanon as a top travel destination, and Orchestrated a multi-channel marketing strategy and collaborated with diverse stakeholders to drive engagement and stimulate economic growth through immersive tourism experiences.
Detailed Description: Click here

Huawei ICT-AI [Team Scoltia]

February 2024 · Huawei (MENA)

Idea: Language translation and voice generation have seen significant advancements, but current systems often lack the ability to capture the subtleties of human speech. This project proposes a solution that combines transformer-based language translation models with voice cloning techniques to deliver a more immersive and authentic translation experience. By utilizing the power of cloud-based AI platforms, real-time audio interception, and deep learning algorithms, our system aims to generate voiceovers that mirror the speaker's tone, mood, and pitch in the target language. With a user- friendly interface and a focus on maintaining privacy and ethical standards, our solution addresses the limitations of existing automatic translation systems and strives to provide a professional and engaging communication tool.
Detailed Description: Click here

NeurIPS 2023 Machine Unlearning

Novemeber 2023 · Kaggle

Idea: The main objective of this competition is to develop an algorithm that takes an AI model and a forget set, part of the training set, leaving behind the retain set, and returns an AI model that’s as closely accurate as the original given model, but the effect of the forget set will be deprecated so that its influence is almost null, meaning that the accuracy on the forget set should approach zero.
Detailed Description: Click here

Website YMY Management System

A business management website for YMY, encapsulating their management processes and automated data entry with Twilio API for WhatsApp Business. It covers a rich backend that

Website BioGenie

A nice website with an easy user interface, and a playful and interactive frontend design, integrated with a chatbot using OpenAI API that helps assess Lebanese LS students’s biology knowledge by answering MCQ questions, provoding all the necessary materials to improve.

Website Musical Instruments HTML views

A basic freelance HTML page for a fundamentals course final based in Canada. The aim was to grasp most of the HTML features.

Website Nollini (discontinued)

A simple flask app for a client incorporating state of the art security best practises to manage a vet owned by the vet owner (the client).

Website Vet Manager

An Ecommerce static website that hooks a target market of Lebanese employees based in Lebanon with copy-A European distributers through submitting and showcasing their products on one central hub.

Website Blood Donnation Website

A simple freelance blood donnation flask app for a software engineering course. Implements required functionalities learned through the course and advanced user authentication and security standards.

AI Image Recognition with CV2

Freelance computer vision project. OpenCV2 script to detect matches based on a HAAR Cascade fine tuned model, with the fine tuning dataset being normal images of matchboxes.

Data Science Parkinsons Disease Data Acquisition and Preprocessing

Helped with devising a PhD student’s research on parkinsons’ disease, by providing state-of-the-art datasets for AI training freely available on the internet, for the follwing data types ‘EEG’, ‘EMG’, ‘X and Y for leg movement’, ‘voice’, and ‘Handwriting’

GUI Python GUI Immigration Management System

My first freelance project. Within a deadline of 3 days, I got a university student’s request to create a GUI application with TKinter, and i accepted with no prior knowledge. Within 2 days, the GUI application was complete and up to standards in regard to security and database management. This demonstrates my willingness to take risks and eagerness to learn new things and ability to manage my time and flexibility with adaptation and working in an agile development environment.

AI Hospital Management System

In a team of 2, we are creating a hospital management system that links clients to doctors and helps them classify their cases by entering their syndromes and predicting with a Random Forrest model the disease they are suffering from, which is my main responsibilty in this project. You can think about it as a social media platform, more of a LinkedIn, for the medical sector, that fosters the interaction between clients and workers in the health field.

Data Science Seismic Waves to Marine Pulses: A Curation Pipeline for Building an Earth Sciences and Biodiversity Data Lake in the Portuguese Caravela Jellyfish and Seismology Studies

This thesis is part of the Data and Intelligence for Smart Systems curriculum at Universit´e Claude Bernard Lyon 1. Conducted under the supervision of Genoveva Vargas-Solar and J´erˆome Darmont, it aims to integrate seismic wave data and marine pulse information. The motivation is to enhance understanding of the inter- play between geological and biological processes, informing scientific discoveries and environmental conservation strategies. By establishing a curated data lake, this research emphasizes the importance of innovative data management and analysis techniques in interdisciplinary projects.

AI PatchNet: A Hybrid Model for Fine Segmentation of MRI Images across Multiple Organs

This experiment is one of 4 datasets on biomedical image segmentation, to train 7 Deep Learning (CNN-based and Transformer-based) Image Segmentation Models, incluidng the new Patchnet Architecture. Abstract: In this paper, we introduce PatchNet, a novel hybrid model tailored for fine segmentation tasks in medical imaging, particularly focusing on MRI scans of various organs. Fine segmentation is essential for accurately distinguishing between different tissue types, identifying small abnormalities, and delineating intricate anatomical structures—key in applications requiring high precision.

Cafeteria Madness

The player does not need a controller to play. An AI model can scan his hand gestrues and map regular patterns to in game player-environment interactions. Demonstrated in the game ‘Cafeteria Calamity’. A team of 4, we made everything, from art to music to code.

Recipe Sharing Platform

The project entails the development of a recipe sharing platform that aims to simplify the process of meal planning and recipe discovery for users. This platform can potentially be extended to mobile devices, catering to a wider audience and enhancing accessibility. The primary objective is to provide users with a user-friendly interface where they can easily search for recipes based on ingredients, preferences, dietary requirements, and available kitchen equipment. Additionally, the platform will include features for recommndations uisng AI recommender systems and tracking health and nutrition, such as calorie intake monitoring, through an intuitive dashboard design.

Better Moodle

Moodle platform can be quite challanging to use, and this difficulty becomes a real issue that all BAU students suffer from during online course reistration. The issue is that the servers at the university are outdated and cannot handle the traffic. Thus, Better Moodle is a mobile flutter app that provides utilities to work aroudn this issue, such as regular caches from the moodle website, like schedule and course offering, notifications when anything is posted on moodle website, and automation services like automated registration with course code list input.

Cifar10 Resnet18 Image Classification

Training an AI model to classify images from the Cifar10 dataset with Resnet18 architecture.

ARC website

ARC needed automated processing and various other features like information and blogs and links and stories and media etc… You get it, just what any club needs, a centralized server running managing the process of running the ARC club fully.

Gym Management System

A Solo project i made to learn Django and Flask, and web development in general. This website manages the gym operations fully, from its employees to its clients. To employees, it is a business management system, but to the users it is simply an ecomerce/dashboard analytics web application.

NVIDIA Canvas API with GUI

The university owns a super computer capable of running the NVIDIA Canvas app, but most students weren’t allowed in the lab without priviliges like being a computer engineering student. Thus, in a due, me and my best friend created an API that links university students on the same network as the Engineering Faculty Labs to the NVIDIA Canvas application found on the super computer, through a Flask-based HTML canvas web app that runs client-side on each student’s device.

CLI Java Password Manager

Created a XOR and rotating key based encryption algorithm that decreases probability of cracking the password exponentially in powers of 2, and implemented it in a nice looking, ascii-art based CLI application that is built with Java so it would be available on any device. The password database is even made with a custom .dat file to avoid sneaky decryptions from hackers.

JavaScript Learning Series

This is a series of projects done with peers in the aim of learning JavaScript. So it is true that these projects are relatively simple and small scaled, but they aren’t just a bunch of copied and pasted tutorials, as there is a little whiff of creativity and imagination in each single one of them.

Java CLI MinMax Game (Discontinued)

Created, in a team of 3, a CLI Java game that utliizes the minmax algorithm to play a board game with a human player

Min Lanes Java CLI Game

This is a little game made during Murex interview prep to boost my DSA Data Structures and ALgorithms knowledge, the objective is to find the minimum lane switches to get to the other side. The CLI is designed to where you can easily visualise the lanes and understand the minimum route chosen by the algorithm.

Telecom Compnay Management DBSM - Website

An SQL Database Project intended to manage a telecom company’s workflow, like Alfa and Touch, then connected to a backend of a website made with Flask for ease and quick of development, just for demonstration purposes, but the website sill demonstrates high standards of security and depth and user experience and UI design.

Iris Classification: A Comparative Analysis

EDA, Model Comparison, And more, for the Iris Dataset. Tested a variety of models and preprocessing techniques to get the better most machine learning results from this dataset.

House Price Regression: A Comparative Analysis

An Overhaul of AI and Data Science, all the techniques you can think of, for the house price prediction problem. Tested a variety of models and preprocessing techniques to get the better most machine learning results from this dataset.

IMBD Sentiment Analysis Using Logistic Regression

Logistic Regression on imbd dataset with binarry classification task, with some exploratory data analysis and detailed preprocessing methods. Developed a really low computationally demanding method to analyse sentiment, demonstrated by imbd dataset.

Robot Mania

In a planned ARC event, an obstacle course is randomly generated, and 2 teams are supposed to compete against one another by building a robot and training its movements using reinforcement learning, or by hardcoding its movements, so that the controls are managed by ROS2 operating system that will run the Genetic Algorithm on both team’s robots and attempt to pass the obstacle course fair and square. This experience is proposed to be gamified using Unity and published on itch.io.

Water Waste Management

In a collab between ARC and the Chemical Engieering Student Society, our project aim is to implement state-of-the-art techniques for water waste management systems using embedded systems and microcontrollers like PIC61F877A with the appropriate sensors provided by the chemical students, based on the environment variables and project scope and budget. It was discontinued however due to lack of interest from the students in the clubs.

Brain Music

Imagine this: You wear a hemlet, and everyone aroud you would hear music from your head! The project is an EEG-scanner that interprets EEG signals into music through MIDI files. It was proposed but rejected since it does not have academic importance in the field of engineering, as compared to a mental health application of the same helmet.

Electronically Encrypted Embedded Systems

Reverse Engineers can extract military secret level code from embedded systems found on military weapons left in the battle field. Thus, We can devise, using convolution, an electronically encrypted embedded system that would explode if unauthorised access was attempted.

Smart Transportation System with Progressive Web Application

See my blog post for a detailed presentation about the project description.

IOT Management Framework

I proposed a cloud-based servers-as-a-service business framework, which was discontinued since i found this business that already implemented my idea. Check it out for more details.

Log Analysis using AI: Random Forrest for user classification behind each log

Hackers and cyber criminals, even normal criminals, are using the internet everyday. An international station like Google and the CIA is more than capable of profiling people based on their network traffic and server logs from around the world. This project aims to use Random Forrests to detect anomalies in analysed server logs and trace it back to a pool of profiled users who could have caused this anomaly, which is the crime. The users however are none other than clusters of logs.

Sound Generator Rolling Machine

Check the PDF for a detailed description of this project proposal. The idea is that everything in the universe has a unique sound, and that is a direct consequence of chaos theory. We see that demonstrated in this project. We can extract the sound of every touchable object in the universe. Each object has a unique surface distribution of atoms, and using a piezo electric material, we can track the distribution of these atoms on the surface, or, if not the atoms, then the unique surface structure to that material. Then, since we have an electric signal, what we can do is infinitely repeat it many times bu rolling the piezo electric material over and over again at the material surface to generate loops in the signal, thus extracting music as verified by fourier transform.

Emotion Detection Using Multimodal AI Models Ensemble

An Attempt to understand and experiment with emotion detection in the most efficient way in every unique combination of AI models and emotion models and data modality. This research is still in its beginning phases, but check this paper that was not published due to its lacking nature.

Symbolic Logic Based AI Architecture with Environmental Integration

An Attempt to understand and experiment with human intelligence and model it using computational models of logic and emotions and whatnot. This is still in a very early stage, and is pending, awaiting the Emotion Generation experiment. The core idea is that every intelligent agent must have a central controlling unit that interacts with its environment and its internal components, facilitating its growth and survival optimally.

Emotion Generation via Adversarial Network

Once the Emotion Detection Model Enseble is complete, we can mimic emotion generation using an Adversarial Network. This research is still in its early phases as it’s awaiting the results of another paper.

From Relative to Real: A Vision Model for Absolute 3D Understanding

This project proposes a novel vision model that transcends relative depth perception and strives towards absolute 3D scene understanding. The issue with our human vision is that we perceive depth with 2 dimensional images in our head at all times. We want to be able to see objects in their true size no matter how far away they are, so identical objects will appear identical at a distance from one another. Moreover, distances on the 3 spacial axis must still persist and appear.

AI Avatar Creator

In our meta-world, humans are relying more on their meta-personas to define their identities than ever before. This means that soon, we will need IDs on the internet, similar to the national IDs we physically hold, which is probably the area of block chain technology. However, assuming that such a secret private key exists, which defines your presence online, then we can create an AI that will scan your face and be able to detect your face forever more, then generate this secret key and integrate it for you in an online identity, so you can for example embed your secret key in any game and get a skin that looks like your meta-self. This is the avatar generator I proposed.

DNA Based Processor

DNA is used to store data in state-of-the-art research. However, what people overlook is that DNA strands have encoded information (ACTG => ACUG) which are then decoded by mRNA and thus generate a sequence of instructions based on codons, tripplets of information (AUCG), and a sequence of these codons create a certain instruction. And so, an instruction set is defined and complex instructions can arise. This has the potential to become an interactive game where your goal is to design the perfect DNA of a living micro organism and understand the intricacies of life itself, or a hobbiest project that is the DNA based Microprocessor.

Machine Unlearning: A Comparative Analysis

This paper explains everything in depth. Published in conference EAAAI 2024, Greece.

Roseman Emotion Classifier

Emotion Detection according to roseman’s cognitive theory of emotions. To run the script, first of all open a terminal in the src/ directory, then run data_preprocessing.py then appraisal_model.py then emotion_model.py then predict.py in that order. Enter the sentence you want to predict in the predict.py to get a prediction.

Novel Arabic NLP Sentiment Analysis

This paper presents a comprehensive study on sentiment analysis in the Arabic language, focusing on five categorical emotional dimensions. We introduce a novel Arabic corpus specifically designed for advanced sentiment analysis. We detail the specifics of a data augmentation technique that inverts the logic of sentences which reverses the predicted emotion, a detailed comparison of state-of-the-art NLP techniques such as TF-IDF, LDA, Bag of Words, etc. We trained machine learning (ML) and deep learning (DL) models for one-vs-many emotion classification and explored the effectiveness of an ensemble model combining these individual classifiers which scored high accuracy values. Furthermore, we analyze unjust voting system inherent in ensemble methods. Our approach highlights the challenges and opportunities in sentiment analysis, and our experimental framework encompasses data preprocessing, exploratory data analysis (EDA), and model training pipelining.

• Video
• YouTube Channel
• BandLab Project
• 1.mp3
• 2.mp3
• 3.mp3