Myrzakhan Naukhanov

Machine Learning Engineer

JSC National Information Technologies

Built production RAG and LLM serving systems, including semantic search and evaluation; reported 40% faster query resolution and 15% better answer relevance.

• Designed and deployed RAG pipelines integrating corporate data with LLMs, reducing query resolution time by ~40%. • Configured inference servers (Ollama, vLLM) handling 500+ concurrent requests with sub-second latency. • Integrated PostgreSQL + pgvector for semantic search across 100K+ documents. • Built retrieval evaluation frameworks (precision@k, MRR), improving answer relevance by 15%. • Designed whole flow of chat-bot with differnt pipelines, including: llm guardrails, intent routing, chat history with chat context window with Redis.

Machine Learning Engineer

Private Institution ISSAI

Fine-tuned multimodal models, trained large models on GPU clusters, and improved KazLLM de-identification NER F1 by 12%.

• Fine-tuned multimodal InternVL (LLaMA/Qwen + InternViT), achieving 7% accuracy gain on internal benchmarks. • Trained 10B+ parameter models on multi-GPU clusters (A100, H100); built scraping pipelines collecting 500K+ samples. • Optimized de-identification workflows for 1M+ records for KazLLM; improved NER F1-score by 12%.

Machine Learning Engineer (CV)

BIOMETRIC.VISION

Built document-processing and biometric verification systems, with 98%+ document accuracy and a 60% reduction in spoof acceptance.

• Built document processing pipeline (OCR + detection + face recognition), handling 10K+ documents daily at 98%+ accuracy. • Reduced spoof acceptance rate by 60% via fine-tuned active/passive liveness detection models. • Led 5 annotators, built custom tools, and improved label consistency by 25%.

M.Sc., Applied Mathematics

Nazarbayev University

B.Sc., Pure Mathematics

Suleyman Demirel University