Generative AI Engineer (LLM)
VG Recruiting Agency
Yerevan | Full time
Job description
EPAM is seeking an experienced Generative AI Engineer to take ownership of improving and optimizing the platform's language models, covering the entire process from fine-tuning and evaluation to deployment and ongoing performance monitoring.
The successful candidate will work closely with Back-End Engineers responsible for serving infrastructure, as well as Data Analysts and business stakeholders who define the quality standards and business objectives for AI-generated recommendations.
This is a fully remote position, offering the flexibility to work from anywhere in Armenia, including from home or from the company's well-equipped offices in Yerevan or Gyumri.
Job responsibilities
- Perform supervised fine-tuning (SFT), parameter-efficient fine-tuning (PEFT), and model adaptation techniques, including LoRA, QLoRA, and adapters.
- Apply preference-based alignment methods such as DPO, ORPO, and RLHF-style approaches to improve model tone, output structure, and domain-specific reasoning for retail next-best-action and content-compliance tasks.
- Manage the entire dataset lifecycle, including sourcing and curating training examples from platform data and expert feedback, developing annotation and review workflows, deduplication, and preventing data contamination across evaluation sets.
- Generate and filter synthetic data and maintain version control for all datasets used in training.
- Optimize model costs and response times through distillation, quantization, batching, caching strategies, and model routing while maintaining agreed quality standards.
- Package and deploy fine-tuned models using Vertex AI or equivalent platforms, and manage model and prompt versioning.
- Monitor production models for performance drift and quality degradation, establishing retraining triggers and rollback procedures.
- Implement and optimize safety measures, scope restrictions, and Personally Identifiable Information (PII) controls for enterprise commercial data, including refusal mechanisms and output validation.
- Collaborate with commercial teams to translate qualitative feedback into actionable training data and measurable evaluation criteria.
Required qualifications
- At least 5 years of hands-on experience with Large Language Model (LLM) fine-tuning, including Supervised Fine-Tuning (SFT) and Parameter-Efficient Fine-Tuning (PEFT/LoRA). Experience with preference optimization methods such as DPO is highly desirable.
- At least 1 year of relevant leadership experience.
- Proficiency in Hugging Face Transformers, PEFT, TRL, Axolotl, or Unsloth, or experience with managed model-tuning services such as Vertex AI, Amazon Bedrock, or OpenAI.
- Strong understanding of model training fundamentals, including tokenization, context length, sequence packing, learning-rate selection, and hyperparameter optimization.
- Knowledge of catastrophic forgetting, overfitting on small domain-specific datasets, and GPU/TPU memory limitations.
- Experience in LLM evaluation, including designing evaluation datasets and metrics for generative tasks, with the ability to assess model performance beyond standard benchmark scores.
- Strong data curation skills, with an emphasis on dataset quality, labeling guidelines, inter-annotator agreement, and data leakage prevention.
- Familiarity with synthetic data generation challenges and filtering techniques.
- Experience with LLM Operations (LLMOps), including model and prompt versioning, drift monitoring, and retraining lifecycle management.
- Knowledge of Google Cloud Platform (GCP), particularly Vertex AI, Google Cloud Storage (GCS), and Cloud Run.
- Experience with Git-based development workflows.
- Proficiency in Python, including working with notebooks, training scripts, evaluation pipelines, and adapting existing framework code.
- Excellent written and spoken English, at B2 level or higher.
Additional advantages
- Experience with Reinforcement Learning from Human Feedback (RLHF) and preference-based model alignment techniques.
- Familiarity with model distillation and quantization methods, including GPTQ, AWQ, GGUF, and bitsandbytes, and understanding their impact on model quality and performance.
- Knowledge of LLM inference frameworks and serving technologies such as vLLM, TGI, and Vertex AI endpoints, including streaming, latency, and throughput considerations.
- Understanding of LLM security practices, including safety guardrails, scope controls, and PII protection for enterprise commercial data.
Benefits
Work Environment and Flexibility
- Opportunity to work on innovative solutions for industry-leading companies and contribute to projects with global impact.
- Flexible working environment, with the option to work remotely or from the company's offices.
- Opportunity to work abroad for up to two months per year.
- Relocation opportunities across the company's offices in more than 55 countries.
- Corporate and social events.
Professional Growth and Development
- Leadership development, career guidance, soft skills training, and well-being programmes.
- Professional certifications, including GCP, Azure, and AWS.
- Unlimited access to EPAM's internal learning platform.
- Free English classes with certified teachers.
Additional Benefits
- Participation in the Employee Stock Purchase Plan.
- Monetary bonuses through the employee referral programme.
- Comprehensive medical and family care package.
- Four additional trust days per year for personal needs.
- Fitness club discounts.
- Additional discounts on hotels, restaurants, stores, and various services.
Contact information
Interested candidates are encouraged to apply here.
Please clearly mention that you learned about this opportunity through the VG Recruiting Agency Telegram channel.