Data Scientist

Topkapı Danışmanlık Elektronik Hizmetleri Pazarlama ve Ticaret A.Ş., a subsidiary of İşbank and operating in Istanbul. Contributing to the development of the online shopping sector in Turkey with the marketplace model that aims to bring together companies of all sizes operating in Turkey through Pazarama.com, to provide an improved experience supported by secure new generation payment solutions by processing customer data in the most accurate way, to meet the needs of consumers.

• Working on search engine with ElasticSearch and implementing semantic search method using word embeddings.
• Building dynamic and static recommendation engines for customers while shopping and visiting the platform.
• Trained Sasrec and Ssept models for sequential recommendation and trained fasttext model based on the semantic closeness of the products in view and wishlist products.

Huawei Technologies Co., Ltd. is a Chinese multinational technology corporation headquartered in Shenzhen, Guangdong, China. It designs, develops and sells telecommunications equipment, consumer electronics and various smart devices.

• Working on user intent extraction from search queries on Huawei App Gallery and using the mBERT model to classify searched query's category in three levels.
• Clustering queries in Turkish, Russian, and other required languages to annotate them, and train a classification model to extract information on which categories users search to display better results.
• Participated in a research paper project on topic-based text clustering. Experiment with different clustering algorithms on different datasets to evaluate and obtain a benchmark.
• Research on Hierarchical Superordinate Word Representations project for improving word vectors and semantic similarities between sub-categories and upper classes.

Etiya is a leading software company providing customer experience-focused and AI-driven Digital Transformation with its award-winning portfolio. It enables rapid transformation, immediate revenue growth, and competitive advantage to companies with its AI-supported products, and microservice-based architecture by bringing agility and flexibility into their business.

• Built a semantic document search engine service for Akbank, one of the biggest banks in Turkey, to reach any document in the bank's database, using word embeddings. Utilised ELK stack on docker, ElasticSearch for indexing the data and using cosine similarity feature to find the semantic similarity between searched queries and document's content, title, and description.
• Trained clustering model for recommendation engine in addition to using word embeddings. Implemented sample user log data and trained word2vec model with the features of products and dates in order for each customer and represented this feature with word vectors. Clustered all customers using the k-means model to assign new users based on their product search and recommends products accordingly. Besides, he adjusted this model to recommend statistically when the user comes for shopping after a session and dynamically by changing the recommended products while the user adds products to their carts. This model was deployed and bought by Basbug Otomotiv one of the biggest car spare parts seller stores.

Xinapse provides new experiences and values of AI communication that speak, listen, read and understand so that humans can interact with each other through artificial intelligence through generative AI technology.

• Worked on building a question & answer system for English and Korean languages using BERT to replace it with the rule-based system that had been implemented inside the company’s chatbot architecture.
• Created a Korean news data crawler from the Naver news site using EFK(ElasticSearch + Fluentd + Kibana) stack inside the Docker container.
• Implemented a more efficient tokenizing method for Hangeul(Korean) named Sentencepiece where originally Wordpiece tokenization is implemented inside BERT. (Conducted by Korea Industrial Research Institute)
• He also built a forecasting business cycle indexes model based on time series Korean news data for a specific industry. Implemented Topic Modeling with NMF, semantic analysis of news data, dynamic factor model, latent threshold model, etc.