Twinby 3 часа назад
Team Lead ML (RecSys)
- Удалённо
PythonMLClickHouseA/B testingAirflowPyTorchDeep LearningNLPMonetisationRecSysTrust & Safety
Описание
Responsibilities
- Own the recommendation system end-to-end: candidate generation, ranking, re-ranking, cold start, exploration for a two-sided dating product where the output must satisfy both parties.
- Be responsible for online results: goal and metrics, quality and drift monitoring, rollbacks and degradation analysis; A/B tests are conducted with a product analyst who is responsible for test design and statistics.
- Take on related ML projects: Trust & Safety (bots and farms, fraud, moderation, liveness), monetization and retention (churn, uplift, personalized offers), NLP for chats (toxicity, spam, fraudulent scenarios).
- Build a team and process.
- Work with the data platform and analytics.
- Write code yourself where it provides maximum leverage: design and key models, critical pipelines, reviews.
Requirements
- Experience with production recommendation systems: system in production for which you were responsible for online results (feed, ranking, catalog, matching).
- Deep learning in practice: PyTorch (or equivalent), neural network ranking and retrieval (two-tower, sequence models, multi-task), training on large data, inference optimization.
- NLP: modern models for text (transformers, embeddings, fine-tuning), classification and ranking based on user text - toxicity, spam, fraudulent scenarios, similarity search.
- Classic ML alongside DL: gradient boosting (CatBoost / LightGBM), features, evaluation; ability to choose a simple solution when a deep one is not needed.
- Confident Python, ability to write and deploy a model to production; understanding why offline metrics do not equal online effect (biases, feedback loop, drift).
- Instinct for data trust: monitoring, reproducibility, versioning, data tests.
Nice to have
- Leadership experience: led people (formally or as a tech lead), hired, grew, built processes.
- Experience with online experiments for recommendations (interference, long-term effects, power) - test design is handled by the product analyst on our side, but understanding is valuable.
- Reciprocal recommendations, two-sided markets (job platforms, marketplaces, taxis, dating).
- Experience in Trust & Safety / anti-fraud (bots, fraud, moderation, liveness), computer vision.
- Monetization experience: uplift, churn, offer bandits.
- Experience with ClickHouse, Airflow, vector search.
- Publications, presentations, mentoring.
We offer
- Real ownership: the recommendation system and ML portfolio of the product are yours.
- Opportunity to build ML from a solid foundation: first trusted data and metrics, then models.
- Autonomy in how the ML process and team are structured; you discuss goals and priorities with the Head of Data.
- Rich dating context: reciprocity, gender imbalance, safety of real people, chat, monetization.
- Employment under the Labor Code of the Russian Federation.
- Work format - remote.