Vladislav Zapromyotov

Machine Learning Engineer

Kaspi.kz

Kaspi.kz is a major Kazakhstan-based fintech and e-commerce company, known for its widely used Kaspi Super App.

Built ML services for document generation, call quality control, and classification at Kaspi.kz, reporting improvements in manual effort, productivity, parsing accuracy, and F1-score.

• Automated the document generation pipeline by building a FastAPI-based microservice integrated with a Milvus vector store and BGE Reranker, reducing manual human effort by 48%. • Architected a real-time call center QC microservice ecosystem using CatBoost and fine-tuned DeBerta, boosting managerial productivity by 20%. • Maximized LLM-parsing of client call scripts from 75% to 95% by improving prompting and leveraging Pydantic validation along with the Constrained Decoding method.

Machine Learning Intern

Software Analysis and Developer Support Lab

Worked on evaluating LLM-generated Android UI tests, reducing model calls and porting DroidAgent function calling to Gemini and Mistral APIs.

• Constructed a baseline functionality coverage for LLM-generated UI units tests for commonly used Android applications. • Improved efficiency by reducing model calls using Custom Structured Outputs construct. • Eliminated prototyping costs by porting DroidAgent framework’s function calling from OpenAI to Gemini and Mistral APIs. • Identified 3 new evaluation metrics to improve quality of generated unit tests, including mutation testing, generated test relevancy, and crash detection.

Bachelor of Science, Computer Engineering

New York University