We are building a National Large Language Model (LLM) for the Uzbek language from scratch — from developing data pipelines to fine-tuning and production deployment. We are looking for a Data Engineer to take ownership of the full data lifecycle for the LLM: collection, cleaning, validation, feeding into training pipelines, and supporting RAG systems.

Responsibilities:

• Pipeline Development: Build scalable data pipelines for training and evaluating LLMs, as well as for RAG systems.

• Data Processing: Collect, clean, normalize, deduplicate, cluster, and validate massive text datasets (news, books, web content, legal documents, conversational speech).

• Quality Assurance: Develop automated data quality assurance systems (filters, heuristics, and ML-based validation methods).

• Tooling: Create and support tools for data annotation, verification, and curation.

• Dataset Preparation: Prepare datasets for:

  • Instruction tuning;

  • Supervised fine-tuning (SFT);

  • "Question-Answer" pair generation;

  • Translation and summarization tasks.

• Collaboration: Collaborate with ML Engineers on tokenizer training, LoRA/QLoRA tuning, data versioning, and experiment tracking.

• RAG Implementation: Develop RAG pipelines (document ingestion, chunking, vectorization, retrieval) and integrate custom inference solutions into production.

• Optimization: Configure workflows to maximize GPU efficiency during training and inference.

Requirements:

• Python: Deep knowledge of Python for data processing and building ML pipelines.

• ML Stack: Solid experience with PyTorch and HuggingFace Transformers, plus a basic understanding of CUDA (sufficient for collaboration with the ML team).

• LLM Expertise: Practical experience preparing datasets for LLMs, working with Instruction tuning and SFT.

• Frameworks: Experience with LangChain or LlamaIndex; understanding of LoRA / QLoRA and tokenizer training processes.

• DevOps Tools: Proficiency in Git and Docker.

• Backend: Experience with backend frameworks (FastAPI or similar).

Conditions:

• Schedule: 5 days a week, from 09:00 to 18:00.

• Employment: Official employment in accordance with the Labor Code of the Republic of Uzbekistan, providing 28 calendar days of vacation.

• Dress Code: No strict dress code — we aim to break stereotypes about government-related organizations.

• Team: Work within a strong team of professionals ready to share knowledge and experience.

• Impact: Participation in large-scale and significant projects aimed at creating services to improve the population's quality of life and optimize business processes in the country's leading enterprises.

• Autonomy: Wide opportunities for independent decision-making and active influence on the company's development.