ETL for Processing Personal Data of Employees

Modern times require modern solutions especially when concerning routine tasks, which can be automated. Lately, deep learning (DL) approaches are becoming more popular to solve some of such tasks, and as these techniques are becoming more sophisticated, they can solve more complex tasks. Some DL models may even surpass humans in detecting objects, correcting errors in text, recognising sentiments in the text or providing answers to all sorts of questions. Moreover, some of the models operate faster than human brain and can handle large amounts of information. In this thesis paper we provide the results of the project dedicated to personal data form processing. The project itself is a ETL (Extract, Transform, Load) process, which uses several fine-tuned deep learning models to correct spelling, detect and order the elements of names and addresses, as well as conventional rule-based methods to correct dates format. To fine tune the models, we generated datasets resembling the contents of the form. As the result of the project, we have developed an ETL for processing personal data provided on MS Excel forms in Russian language. The ETL combines both rule-based and deep learning approaches to solve formatting and spellchecking tasks.