Why this role exists
DealSeek has more inventory than any person could browse. As we grow and add more retailers, ranking and recommendation will only become more important. Our recommendation system decides whether a shopper finds something genuinely useful or scrolls past noise. We have the users: our A/B tests are decisive within hours, and we have the privilege of no cold-start problem. Now we need a skilled engineer to tinker, deploy better models, and make every session more relevant.
What you'll do
- Own the full ML lifecycle: data analysis, feature engineering, training, evaluation, deployment, monitoring, and iteration
- Build ranking and recommendation models with Python and LightGBM that score millions of deals against real production traffic
- Design A/B tests and causal measurement strategies that connect offline metrics to shopper behavior
- Use embeddings and Milvus vector search to improve discovery, personalization, and cold start
- Monitor model quality, drift, reliability, serving latency, and scalability in production
- Use ClickHouse and Postgres to turn behavioral and catalog data into useful features
- Partner with backend engineers on fast, observable model serving in Go
- Improve our experimentation workflow with Optuna and Weights & Biases
Qualifications
- Hands-on experience shipping ranking, recommender, search, or ads models
- Strong Python and practical experience with boosted trees, embeddings, and vector search
- Experience owning models beyond the notebook, including deployment, monitoring, and iteration
- Strong judgment about A/B testing, causal measurement, and the limits of offline evaluation
- Ability to read production Go and collaborate on model-serving systems
- A bias towards iteration and measurement over a single, long-term research project