IT company GNIVC is a partner of state companies and leaders of Russian business, a developer and system integrator of the largest state information systems, as well as commercial solutions for tax monitoring.

• The company is among the TOP-100 best employers in the country and ranks 9th in the "IT companies" category for 2025 according to the hh.ru employer rating among large companies;
• We are in the top 25% for happiness level among IT companies in Russia for 2025 according to Happy Job;
• We have an AI sandbox – an environment for experiments and real projects on modern open-source large language models. Here you can apply AI to optimize your work, automate processes, and implement your own ideas from hypothesis to result;
• We are an accredited IT company.

Tasks:

• development and refinement of a classifier for product names based on BERT architectures (including pre-training, fine-tuning, and model optimization);
• organization and control of the data labeling process: setting tasks for labelers, quality control, pipeline automation;
• dataset preparation: data cleaning, normalization using pandas, datasets (Hugging Face), and regex;
• model architecture design and implementation: experiments with BERT, DistilBERT, custom heads, ensemble methods;
• monitoring model performance in production: quality metrics, drift detection, A/B testing, automated fine-tuning;
• working with PostgreSQL: creating tables, writing stored procedures and functions, query optimization (indexes, materialized views, EXPLAIN ANALYZE), ETL pipelines for model data;
• model integration into production: Docker containerization, GPU/CPU monitoring.

Requirements:

• higher education (preferably in computer science, mathematics, or related fields);
• 2+ years of experience in NLP/ML, including fine-tuning transformer models (BERT/RoBERTa/DistilBERT);
• deep experience with PyTorch, Hugging Face Transformers, pandas, datasets;
• experience with PostgreSQL: SQL, schema creation/optimization, stored functions, query performance;
• knowledge of model optimization techniques: quantization, layer-wise LR, custom loss functions;
• production ML experience: monitoring, anomaly detection, model serving;
• strong command of Python, Git, Linux/Shell scripting.

Will be a plus:

• experience with text classification (product catalogs, search/recommendations);
• skills in organizing data labeling and data pipeline engineering (ETL, preprocessing);
• experience with ONNX for inference, multi-GPU training (DDP).

We offer:

• flexible work formats: possibility of working in the office, hybrid schedule, or fully remote within the Russian Federation;
• working hours: five-day work week (Mon–Thu from 09:00 to 18:00, Fri from 09:00 to 16:45);
• decent compensation: competitive salary based on interview results, as well as bonuses for effective work and achieved results;
• official employment: full compliance with the Labor Code of the Russian Federation, including paid leave (with an additional payment of 50% of the salary after 11 months of work in the Company) and salary payment twice a month;
• health care:
  • compensation for sick leave up to 7 days with full payment equivalent to a working day,
  • voluntary medical insurance (VMI) after the probation period, with a wide list of leading medical institutions, including quality dentistry,
  • reimbursement of up to 50% of sports expenses;
• development and training:
  • professional training and certification at the company's expense,
  • organization of internal and external meetups, hackathons, conferences, seminars, and trainings,
  • partnership programs for learning foreign languages and developing professional skills from Skyeng and Skillbox,
  • access to the corporate library on the Alpina Digital platform;
• additional days off: possibility to take 5 additional paid days off (wellness days) during the calendar year (from January 1st to December 31st) for employees who have worked in the company for more than 11 months.