Intern AI Engineer
National Information Technologies JSC (NITEC)
Kazakhstan’s national IT company, involved in the development and operation of government information systems and e-government services.
May 2026 — PresentDeveloped and deployed a bilingual RAG chatbot and prepared its knowledge base for production use.
• Developed and deployed a bilingual (Russian/Kazakh) RAG chatbot at NITEC using LangGraph for orchestration and FastAPI, containerized with Docker and deployed to a production server.
• Built the underlying knowledge base by cleaning and merging multiple administrative-procedure and FAQ sources into a structured, embedding-ready schema.
Intern AI Engineer
TTC Transtelecom
A Kazakhstan telecommunications company that also provides communications services for the railway sector.
May 2025 — Aug 2025Worked on RAG-based customer support tools, including a call center bot and a Telegram bot for railway customers.
• Enhanced a call center AI-bot using Retrieval-Augmented Generation (RAG). Researched embedding models to balance performance and computational cost, and experimented with a Cache-Augmented Generation (CAG) alternative to RAG.
• Built a Telegram bot using the Aiogram library in Python, integrating a RAG-based system to support national railway company customers.
Intern Data Scientist
National Information Technologies JSC (NITEC)
Kazakhstan’s national IT company, involved in the development and operation of government information systems and e-government services.
Oct 2024 — Feb 2025Applied causal machine learning and open-source LLMs to resident data, and developed a Kazakh-Russian speech translation program.
• Based on large-scale data of residents segmented by districts, and after performing data preprocessing, I used causal machine learning techniques to identify the most important features, generated recommendations using different open-source LLMs, including KazLLM.
• Developed a program for synchronous translation from Kazakh to Russian/ from Russian to Kazakh (TTS and STT) using open-source speech recognition models. For translation, I experimented with Soyle (both locally and via API) and Tilmash.
Research Assistant
Nazarbayev University
Aug 2025 — PresentResearches LLM fine-tuning for edge deployment and developed a CNN-Transformer-CTC model for continuous sign language recognition.
• Fine-Tuning LLMs on Edge Devices: Implemented and evaluated fine-tuning methods (full FT, LoRA, BitFit) on different LLM architectures (Transformers, SSMs), assessing performance across multiple benchmark datasets using local and cloud hardware.
• Fine-Tuning LLMs on Edge Devices: Analyzed the feasibility of deploying fine-tuned models in resource-constrained local software environments.
• Continuous Sign Language Recognition: Independently conducted end-to-end research, from problem formulation and dataset preparation to model development and evaluation.
