The vacancy is strong in task clarity and requirements but lacks compensation details.
no salary info
Job description
Join Aviasales as an ML Engineer to lead end-to-end ML projects, enhance transformer models, and build classic ML pipelines. Collaborate with teams to deliver solutions and participate in A/B testing design.
Responsibilities
### Responsibilities
- Lead ML projects end-to-end and collaborate with colleagues;
- Fine-tune transformer models for our content classification tasks (intent, topic, monetization);
- Build classic ML pipelines on tabular data: feature engineering, boosting, validation;
- Work with LLM through our gateway: write and iterate prompts, collect eval sets, measure quality;
- Participate in A/B testing design;
- Collaborate with SE to bring solutions to production.
Requirements
### Requirements
- 3+ years of experience in ML/DS with real models in production, not just in notebooks, experience supporting the full lifecycle of models, including monitoring and regular updates;
- Proficient in Python: OOP, testing, typing;
- SQL at a level to independently retrieve data for features and analysis;
- Git and basic understanding of CI/CD;
- Experience with FastAPI and Docker;
- Experience with neural network models, especially for text data: at least fine-tuned BERT-like transformers for your tasks, ideally understand how algorithms work internally and can optimize or modify them if necessary;
- Strong knowledge of classical ML: understanding of main model families and their applicability limits, validation principles, feature engineering, and key metrics for different task classes;
- Experience with LLM API: writing prompts, collecting few-shot examples, measuring quality on eval sets, using LLM-as-a-Judge;
- Understanding of A/B testing — design is covered by analysts, but you should be able to communicate with them effectively.
About Aviasales
Aviasales is a leading travel metasearch engine that allows users to search and compare cheap air tickets, hotels, and car rentals, redirecting them to airlines and agencies without selling tickets directly. It operates primarily in Eastern Europe, CIS, Asia, and other regions, with a monthly audience exceeding 15 million users as of 2021.
