Yerzhan Orazayev

Senior ML Engineer

KazDream Technologies

Architects the platform’s AI/LLM subsystem, agent tooling, observability, and multimodal content pipelines.

• Architect and deliver the AI/LLM subsystem (agentic pipelines, embeddings/search, prompt infrastructure) of a multi-service content-intelligence platform, integrated across its Python/FastAPI, Go, and Airflow/Kafka microservices to analyze Instagram, TikTok, YouTube, Facebook, Telegram, and news content end-to-end. • Built the agent harness behind a dialog agent: typed tool-calling (15+ tools) for multi-step reasoning, RLS-aware NL→SQL execution, and context-window/token-budget management for long-running sessions.

Senior ML Engineer

Forte

Kazakhstan-based commercial bank.

Led LLM workflow automation, built OCR and translation document pipelines, and developed money-mule detection.

• Led LLM-driven workflow optimization across operational departments: customer-support automation, document processing, and internal-tooling copilots. • Built document-processing pipelines combining OCR with LLM-based machine translation, producing structured Word, Excel, and PDF outputs. • Developed anti-fraud dropper (money-mule) detection combining traditional ML classifiers with in-context-learning transformer models.

Middle ML Engineer

2UP Ltd

Built production LLM, RAG, and agent workflows alongside license-plate recognition and video analytics systems.

• Integrated LLMs (OpenAI API and self-hosted open-source models) into production services for text generation, classification, and extraction tasks. • Built RAG pipelines with Qdrant and ChromaDB as vector stores, including embedding generation, chunking strategies, and retrieval evaluation. • Developed agentic workflows with Pydantic-AI using structured outputs and tool-calling for multi-step reasoning. • Built license-plate recognition system combining YOLO-based plate detection with OCR for character extraction.

Junior Machine Learning Engineer

KazDream Technologies

Trained Kazakh TTS models, built ASR data-labeling pipelines, and developed production APIs for ML models.

• Trained Kazakh text-to-speech models (FastPitch2 + WaveGlow vocoder) on professionally recorded speaker datasets; handled data collection, curation, and post-processing. • Built auto-labeling pipelines to enrich training data for ASR (speech-to-text) models. • Worked on NLP tasks and developed backend APIs to serve ML models in production.

PhD, Computer Science

Astana IT University

MSc, Electrical Engineering

KAUST

BEng, Electrical & Electronic Engineering

Nazarbayev University