Portfolio projects
Hey I'm Gal. I've worked as a data scientist and deep learning engineer for the past 5 years and always on the look out for new projects and great people to work with and learn from.
I'm available to connect on LinkedIn or through email.
Over the years I've worked on a broad range of robotics, computer vision, natural language processing, tabular and data science challenges. Most recently on optimization of Convolutional Neural Networks for deployment at split second execution, and before that integrating object detection algorithms in resource scarce underwater robots. I've also helped Fortune 500 companies and Governments identify and catch financial fraud and money laundering. I like to keep busy learning new technologies and solving real world problems. You can see some examples below.
CloudLab: demystifying medical documents
- Built to aid a family during a hospitalization, this app offers real-time access to vital medical data via web and mobile platforms. Utilizing Python's Streamlit, it features GPT-driven querying and data visualization through Pandas and Matplotlib. This solution exemplifies impactful, need-driven tech innovation.
COVID-19 tracker
- A notebook I created in February 2020 to track the growth of a new, deadly virus called COVID 19. At the time, there were two outbreaks, one in Hubei, China and one in Northern Italy. This notebook would run daily, pulling the latest WHO report of figures of infected and deaths, and visualise how the outbreak unfloded in different countries.
Social Robot
- Toibot is a social robotics project I initiated in partnership with Intel Israel and toilab to create an open-source platform that integrates computer vision, speech to text, a chat bot and text to speech on one easy-to-use python platform. Messaging used ROS framework and computer vision based on Intel's Openvino system. Chatbot used some Google Dialogflow funtionality and both S2T and T2S using Google API. You can also see a short demo here.
Text Classification
- Text classification and prediction with a Convolutional Neural Network, a Bidirectional LSTM and a Transformer trained on the 140m strong Amazon review dataset using word vectors. This was actually one of the last tasks on Y-Data's NLP course. It's interesting to compare three 'evolutions' of neural nets - with the final implementation a SOTA distilBert. All networks written in Pytorch.
Rasbperry Pi
- Since moving to Tel Aviv, I've been suffering from continuous tresspassing from my pet dog Shmoops. Every time I left the house I'd would return to smelly sheets. So I created this Raspberry pi based system to notify of the violation and rectify the naughtiness in real time. You can watch it in action here.
CNN for Image Segmentation
- As part of my work with Healthy.io I implemented a convolutional Neural Network for the purpose of segmenting chronic wounds in RGB images. This project was implemented in Tensorflow. This included collecting the dataset, tagging it, training the model on the cloud, testing and deploying. Healthy.io is an amazing company and a fantastic place to work.
Retail Purchase Prediction
- Exploratory Data Analysis (EDA) of an 80k row dataset of consumer behaviour online to predict if a customer will purchase or not. I compare the performance of multiple algorithms and multiple features (num_visits, prodcts_in_cart,target_product_price, age, etc...). The Machine learning models inlcude Random Forest, Linear SVM, Logistic Regression, Multinomial Naive Bayes & XGBoost; all run with 5-fold validation and various imputation strategies.