Search is one of the key drivers of GMV growth. Our team is constantly improving the relevance of search results and the performance of the engine so that users can quickly find what they need.
The scope of tasks includes full-text and vector search for 2 languages simultaneously: Ru and Uz, unified ML ranking for ads and organic results, suggestions, typo correction, and much more.
We have a small ML team of 4 people, so there will be an opportunity to work on different parts of the search stack.
What you will do:
- Unified ranking of organic and advertised products: training models to predict purchase probability in search and catalog, optimizing them for various business targets
- Hybrid retrieval: full-text retrieval - Elasticsearch (morphology, synonyms, training coefficients for boosts) - neural dense retrieval on PyTorch (two-tower encoders, embeddings, ANN)
- Query understanding: search suggestions, models for typo correction, category prediction, NER for entity extraction, LLM for automatic search for problems with popular queries
- Image search: we are just starting to work on this
- Features and data: extracting signals from clickstream and relevance labeling, working with offline metrics that correlate with online metrics later
- Production and reliability: deploying models to production, monitoring quality (online metrics, DQ, alerts).
- Experiments: product analysts will help with design, but an understanding of statistics and the selection of correct metrics is required.
- Engineering culture: MLOps, clean code, reviews, system design, documentation
Our stack:
- ML/DS: Python, NumPy/Pandas, scikit-learn; CatBoost; PyTorch + Hugging Face, ONNX, MLFlow
- Engine: Elasticsearch
- Data and production: ClickHouse, PostgreSQL, Airflow, Spark, Docker/Kubernetes, Kafka
- A/B experiments: GrowthBook
What we expect:
- 4+ years of experience in ML/DS: you have trained and deployed ranking/retrieval models and understand their limitations
- Deep understanding of the search pipeline: sparse/dense retrieval, LTR, offline metrics and their connection to online performance.
- Strong Python + SQL, engineering maturity.
- Experience with A/B experiments: design, launch, analysis, decision-making.
- Good communication skills: you can explain complex things in simple terms and agree on goals/metrics.
It will be a plus if you have 2 out of 3:
- Experience with search/advertising in search
- Experience with recommendation systems
- Experience in the e-commerce domain
Why it's interesting:
- We offer a broad area of responsibility and the opportunity to influence architectural and product decisions. We are also happy to discuss and implement your initiatives
- Unique culture – we have preserved the startup spirit while building mature processes
- We set measurable goals as a team and don't just achieve them "for the sake of it", but measure the effectiveness and overall impact of our work on the business
- Work in teams of strong specialists where depth of expertise and engineering mindset are valued
- Teams listen to and hear each other, acting as partners rather than executors
What we offer:
- Remote work from anywhere in the world or a cozy office in Tashkent
- You can grow in an engineering or management track here, and we have regular performance reviews
- We pay at the level of top companies in the Russian market
- Training and development – we support it both within the company and beyond (meetups, conferences, professional training, publications). We also help develop your personal brand
- Base – a community of professionals with a desire to do great work. A nice bonus – VHI (Voluntary Medical Insurance) tied to your location, training, and other perks
