Experience

Kaspi.kz

• Kazakhstan
• Technology, Information and Internet
• 700 & Above Employee

• Data Scientist

  • Jan 2021 - Present

  - Deployed a high-load BERT-based microservice for an outbound voice bot that classifies client intent, replacing 700 call centre managers and handling nearly 5 million calls per month. - Conducted an A/B test comparing the Rasa DIETClassifier to a custom BERT implementation in Python, demonstrating that the BERT-based classifier made a 7 percentage point reduction in errors from overall calls compared to the Rasa-based solution. - Identified necessary resources for handling production load in the voice bot classifier using JMeter and implemented caching with the Python lru_cache library, thereby enhancing the system's efficiency in managing high load. - Developed a Catboost model to monitor the performance of the voice bot, reducing the error response time from 24 hours to 1 hour. - Constructed a BERT model that identifies rude behaviour from managers towards customers, leading to a 58% reduction in customer complaints within 6 months. - Implemented a Python project that employs the Levenshtein distance algorithm to accurately match and identify partners' names in a database, resulting in a time savings of 8 hours per week for data analysts. Show less