• Owner of user segmentation service which provided a custom experience for over 80 million users.• Increased click-through-rate for segmentation campaigns by 14% and capacity by 50%.• Developed a product recommender model trained daily on more than 600,000 products servicing more than 1700 businesses.• Trained and deployed models to serve more than 100 million views daily.• Increased text classification accuracy by 20%.• Deployed category Detector model in production… Show more • Owner of user segmentation service which provided a custom experience for over 80 million users.• Increased click-through-rate for segmentation campaigns by 14% and capacity by 50%.• Developed a product recommender model trained daily on more than 600,000 products servicing more than 1700 businesses.• Trained and deployed models to serve more than 100 million views daily.• Increased text classification accuracy by 20%.• Deployed category Detector model in production supporting 5 times higher throughput than average load.• Increased subject-based campaign’s income by 50% using a consistent and scalable deployment. Show less

-Led a team to create a chatbot solution for a bank's customer service using g transformer models to answer more than 200 question types with a limited dataset of fewer than 1500 questions.-Increased chatbot accuracy by experimenting, writing, and testing custom NLP modules to take advantage of Persian language models such as BERT.-Created a suitable dataset for training by clustering user data.- Curated chatbot based on needs of stakeholders by holding B2B meetings.-Trained and… Show more -Led a team to create a chatbot solution for a bank's customer service using g transformer models to answer more than 200 question types with a limited dataset of fewer than 1500 questions.-Increased chatbot accuracy by experimenting, writing, and testing custom NLP modules to take advantage of Persian language models such as BERT.-Created a suitable dataset for training by clustering user data.- Curated chatbot based on needs of stakeholders by holding B2B meetings.-Trained and tuned LSTM, CNN, and attention-based neural networks for text-to-speech and speech-to-text in Persian.-Developed tools and scripts for preprocessing the voice dataset to fit the models' input.-Gathered text data and guided a narrator to create a dataset of 14000 voice lines adding to more than 24 hours for TTS.-Made the model production-ready using Docker and Fast-API-Built TensorFlow from base to match the officially unsupported CUDA version required by our GPU with the specific TensorFlow version required by our code.-Mentored more than 20 interns. Show less

- Developing custom user management and accounting software on OpenWRT.- Write efficient backend code for routers' limited process power to enable parents to have control over the family members' usage of the internet. - Write backend code to enable users to control various network and OS settings.- Achieved proficiency in Lua programming- Achieved proficiency in embedded systems programming

teacher assistant for programming fundamentals