Develop LLM-based moderation: translate moderation regulations into model instructions, analyze errors, and improve solution quality
Build reproducible experiments: compare models, prompts, input data composition, and decision thresholds
Apply and develop automatic prompt optimization, fine-tune models and LoRA adapters where justified by quality and cost
Prepare datasets: organize collection and labeling, check data quality and sample representativeness
Integrate solutions into auto-moderation pipelines and ensure stable production operation: consider quality, latency, computational cost, and error risks
Set up quality and degradation monitoring, alerts, and regular improvement iterations
Evaluate the impact of solutions on product and business metrics
Requirements
Ability to conduct the full cycle of ML development: from data and model work to production, monitoring, and impact assessment using A/B tests or equivalent approaches
Extensive experience in NLP or significant experience deploying and supporting models in production
Experience with LLMs in applied tasks and ability to verify solution quality on data
Fine-tuned LLMs and VLMs, worked with LoRA, PEFT, and inference optimization, e.g., with vLLM
Confident Python and DS stack skills: NumPy, pandas, scikit-learn, CatBoost or equivalents, ability to work with DL frameworks, e.g., PyTorch
Understanding of how to choose methods for a task and justify trade-offs between quality, cost, speed, and risks
Ability to assess model quality at launch and track its changes over time: drift, degradations, data issues
Write SQL queries at a level sufficient for data analysis and quality diagnostics
Read English-language technical literature
Nice to have
Experience in moderation, Trust & Safety, or fraud prevention
Developed tools for AI agents, MCP servers, or automated ML processes with them
Familiarity with MLOps practices: Kubeflow, K8s, feature store, MLflow
Configured monitoring and alerts in Grafana, participated in A/B tests and canary releases
Used LLMs in applied tasks
Worked with CV or multimodal tasks, e.g., VLM
We offer
Full-time employment, remote or hybrid (Moscow)
Large-scale engineering process where you can participate in all stages of ML solution work
Interesting development tasks and impact on product and business metrics
Culture of mutual assistance and knowledge sharing
Transparent career track and growth tools: seminars, mentoring, meetups
Extended health care: medical insurance, life insurance, cancer insurance
Psychological support service, as well as financial, legal, and healthy lifestyle consultations for you and your loved ones
Corporate sports: group training, well-being app
Special offers from partners and discounts on Avito services: housing booking, promotion services, mortgage insurance