Machine Learning Scientist
2022 - PRESENTUnited Health Group- Acted as a researcher trying to push the state of the art in NLP applied to healthcare.
- Worked routinely with PyTorch, Hugging Face, NLP, GPT3, S4, Transformer-based models, many other language models (LMs), etc.
- Improved state of the art in healthcare. My role mainly was researching with NLP.
Technologies: Natural Language Processing (NLP), Artificial Intelligence (AI)Data Scientist
2020 - PRESENTDisney Streaming Services- Designed and implemented advanced NLP pipelines for the Disney Streaming Chatbot for customer services.
- Managed text summarization and topic modeling on survey data.
- Deployed a designed chatbot using cloud services.
Technologies: Chatbot Conversation Design, Deep Learning, Natural Language Processing (NLP), TensorFlowMachine Learning Developer
2020 - PRESENTUSC/ISI (Information Science Institute)- Developed machine learning algorithms for event prediction in news corporations, using different technologies, models (e.g., TensorFlow, BERT).
- Turned developed codes into deployable products using containers and Kubernetes.
- Assisted the integration team responsible for delivering a multi-faceted product comprised of different analytics engines.
Technologies: Kubernetes, Deep Learning, Natural Language Processing (NLP)Machine Learning Researcher
2019 - PRESENT3M/MModal- Focused the research on improving NLP pipelines that are used for the summarization of patient-doctor conversations.
- Adapted recent advances in deep learning (applied to the NLP domain) to the company's internal domain to improve the deployed pipeline.
- Worked on a variety of healthcare-related NLP tasks. Used technologies and libraries such as deep learning, NLP, transformers, PyTorch, PyTorch Lightning, Hugging Face, etc.
Technologies: Natural Language Processing (NLP), Deep Learning, PyTorchSenior Machine Learning Engineer
2020 - 2020Liberty Defense- Designed and showcased a deep convolutional neural network for the prediction of threats.
- Involved state-of-the-art image segmentation and detection such as Mask-RCNN for the segmentation of threats (e.g., in cases hiding guns or having guns with them).
- Achieved remarkable accuracy of (95%) in detecting cases carrying guns. Used the TensorFlow library for implementation. Trained on CUDA GPUs.
Technologies: Deep Learning, TensorFlow, Keras, PythonSenior Software Engineer
2018 - 2019Home Depot- Designed and implemented deep models for search and personalization. The task was to rank items returned by a search engine for different search phrases with respect to their relevance and satisfaction of users.
- Trained NLP models implemented with TensorFlow and trained on GPU. Used recurrent neural nets along with siamese networks. Integrated multiple modalities such as user behavior.
- Required preprocessing scripts written in the Spark framework to generate and preprocess large datasets.
Technologies: Deep Learning, TensorFlow, PythonSoftware Engineer Intern
2018 - 2018Verizon Connect- Developed an app using advanced recommender systems for recommending the best matching shopping places for drivers. Used and sorted through a large amount of data accumulated in data clusters.
- Utilized driver behaviors as well as their personalities and demographics to train an integrated deep recommender system. The data was accumulated from a large number of vehicles consuming the product (dongle).
- Integrated two types of recommender systems; the content-based filtering methods, and the collaborative filtering method. Content-based modeled individual personal information, while the collaborative modeled driving behaviors and habits.
- Used TensorFlow and Python to achieve the task using both collaborative and content-based filtering approaches. Trained end-to-end.
- Achieved significant performance in predicting the preference of drivers for their shopping center of interest.
Technologies: Deep Learning, TensorFlow, PythonData Analyst Intern
2015 - 2015UCB Pharma- Developed a deep learning pipeline based on auto-encoders to predict Parkinson's disease from claims data. The goal was to predict whether the person has Parkinson's based on past visits at different doctors.
- Utilized the H2O library in R to implement a deep network from features describing the patient's past medications and diagnosed codes. Achieved an impressive prediction performance of about 90%.
- Identified cases in the early stages of the disease (to receive a more successful treatment), by using the trained model to find trial cases sooner.
Technologies: Deep Learning, R