ml инженер for business automation

The company develops AI solutions aimed at automating real business processes using NLP, NLU, LLM, OCR, STT, TTS, and embedding models. Its focus is on creating, training, fine-tuning, evaluating, and deploying custom ML models rather than merely using ready-made AI services.

• Fine-tune LLM models on domain-specific datasets;
• Develop SFT, LoRA/QLoRA, instruction tuning, and domain adaptation pipelines;
• Create NLP/NLU models for intent classification, entity extraction, text classification, question-answering, and semantic search;
• Create, train, or fine-tune custom OCR and Document Understanding models;
• Develop OCR pipelines for scanned documents, PDFs, forms, tables, and image-based documents;
• Establish text detection, text recognition, layout analysis, and post-processing processes;
• Select and fine-tune embedding models to improve domain-specific retrieval quality;
• Develop chunking, embedding, vector search, hybrid retrieval, and reranking mechanisms for RAG systems;
• Prepare datasets, perform data cleaning and augmentation, and monitor annotation quality;
• Develop metrics, benchmarks, and test datasets for model evaluation;
• Optimize model quality, accuracy, latency, stability, and resource consumption;
• Deploy ML models to production as APIs or microservices;
• Organize training pipelines, experiment tracking, model registry, and deployment processes;
• Integrate ML, NLP, OCR, and LLM solutions with internal systems;
• Participate in technical architecture selection.

• Strong Python knowledge and experience writing production-ready code;
• Deep practical experience with PyTorch;
• Experience with the Hugging Face Transformers ecosystem;
• Practical experience with LLM fine-tuning;
• Knowledge of SFT, LoRA/QLoRA, PEFT, instruction tuning, or domain adaptation;
• Experience training or fine-tuning NLP/NLU models;
• Experience creating or fine-tuning OCR or Document Understanding models;
• Strong understanding of embedding models, semantic search, vector search, and retrieval models;
• Practical experience with RAG, hybrid search, reranking, and vector databases;
• Ability to independently handle dataset preparation, preprocessing, augmentation, and evaluation;
• Ability to manage model training, fine-tuning, evaluation, deployment, and monitoring end to end;
• Experience with Docker, Linux, Git, and FastAPI;
• Understanding of trade-offs between model quality, inference speed, resource consumption, and scalability;
• Nice to have: experience training or fine-tuning models for low-resource languages, STT and TTS experience, GPU training, distributed training or mixed precision training, quantization, batching, caching, pruning, vLLM, TensorRT-LLM or other model serving technologies, CI/CD, production ML monitoring, and OCR/RAG systems for large document collections.

• Opportunity to work on real production projects in AI;
• Work with modern technologies in NLP, LLM, OCR, Embedding, and Voice AI;
• Opportunity to work with large volumes of real data and complex business processes;
• Active participation in technical decision-making;
• Professional team and growth opportunities;
• Salary and work schedule are agreed upon during the interview.