KazakhstanTil-Qazyna National Scientific and Practical Center
AI Engineer
1 нед. назад
KazakhstanCISJuniorOnsite
pytorchhugging facescikit-learndockerlinuxgitlorafastapi
Til-Qazyna is focused on building digital and AI infrastructure for the Kazakh language. They seek an AI Engineer specializing in NLP, corpus engineering, and LLM fine-tuning.
About the company
- Til-Qazyna is a national scientific and practical center dedicated to building the core digital and AI infrastructure for the Kazakh language.
- We develop foundational language resources, linguistic tools, and open-source models, including custom text corpora, morphological analyzers, RAG pipelines, and fine-tuned Large Language Models.
About the product
- Focus on natural language processing, corpus engineering, and LLM fine-tuning for Kazakh language understanding and generation.
Responsibilities
- Fine-tune, evaluate, and benchmark open-source LLMs (Llama, Mistral, Qwen, Gemma) using parameter-efficient methods (LoRA, QLoRA, PEFT).
- Develop, maintain, and optimize rule-based and neural NLP components, including morphological analyzers, lemmatizers, tokenizers, and terminology parsers.
- Design automated pipelines for large-scale text crawling, data cleaning, deduplication, synthetic data generation, and dataset curation.
- Package NLP models and LLMs into containerized microservices (Docker, FastAPI) and optimize inference using serving frameworks (vLLM, Ollama, TensorRT-LLM, ONNX).
Requirements
- 1+ years of hands-on experience as NLP / AI / Machine Learning engineer.
- Solid understanding of the Transformer architecture, attention mechanisms, embeddings, and tokenization.
- Experience with core ML/NLP libraries: PyTorch, Hugging Face (transformers, datasets, accelerate, peft), Scikit-learn.
- Experience building data preprocessing pipelines and working with structured/unstructured text data.
- Practical knowledge of Docker, Linux, and Git.
- Degree in Computer Science, Data Science, Applied Mathematics.
Nice to have
- Hands-on experience with modern LLM serving and optimization engines (vLLM, TGI, llama.cpp, quantization techniques like AWQ/GPTQ).
- Experience with RAG stacks (LangChain, LlamaIndex, ChromaDB, Qdrant, FAISS).
- Understanding of the morphological, agglutinative, or grammatical characteristics of the Kazakh language.
- Familiarity with alignment techniques (DPO, RLHF) or synthetic data generation.
Conditions
- Direct impact on sovereign AI development and language technology used at scale.
- Official employment in full compliance with the Labor Code of the Republic of Kazakhstan.
- Access to high-performance GPU compute clusters for model training and experimentation.