About the Role
We're hiring a Lead Data Scientist/ML Engineer to own and improve Adtima's ads recommendation system - matching the most relevant ads from over 100,000 to 80 million Zalo users daily. Your models directly influence ad revenue and advertiser ROI.
The system covers retrieval, ranking, continuous retraining, and real-time serving (in collaboration with Backend / ML Ops). Your job is to understand, maintain, and develop it further. We welcome DS, ML Engineer, or Applied ML Research backgrounds.
What You'll Do
Model Development & Improvement
- Iterate on ML models across the full pipeline - retrieval, ranking, engagement and conversion prediction. Explore new architectures (transformers, deep interaction networks, knowledge distillation), design better features, and tackle challenges like multi-task prediction, calibration, and cold-start for new campaigns;
- Ensure data quality across the ML lifecycle - validate labels, detect pipeline anomalies, understand sampling biases, and maintain data integrity.
Production Observability
- Build observability that helps the team understand model behavior - prediction distribution, segment-level performance, and performance changes over time;
- Monitor data drift, training-serving skew, and prediction quality. Close the feedback loop between model predictions and real business outcomes;
- Conduct exploratory data analysis on training data, features, and model outputs to uncover patterns, anomalies, and improvement opportunities.
Experimentation
- Leverage the existing A/B testing infrastructure to evaluate model changes - design experiments, analyze results, and make informed deployment decisions.
Technical Direction
- Define what to explore next, what data signals to acquire, and what improvements have the highest impact. Stay current with relevant advances in recommendation systems and deep learning.
Collaborate with:
- Backend / ML Ops (deployment & serving), Data Engineering (pipelines & data availability), Product (business objectives & success metrics), Leadership (model performance visibility & technical direction).
What you will need
Level: Lead (5+ years of relevant experience)
Must-have:
- Deep learning fundamentals + PyTorch (or equivalent). Experience with architectures used in recommendation/ranking (deep interaction networks, two-tower, attention/transformer-based);
- Hands-on experience building recommendation or ranking models in production. Understanding of retrieval → ranking design;
- Feature engineering - embeddings, historical aggregation, cross-feature interactions.
- A/B testing for ML models - experiment design, statistical rigor, online/offline metric alignment;
- Experience building ML observability and monitoring in production;
- Python, SQL, data processing at scale (Spark). ML pipeline orchestration (Airflow or equivalent);
- Clear communication - able to explain model behavior and trade-offs to engineering, product, and business stakeholders.
Nice-to-have:
- Experience in ad tech, e-commerce, marketplace, or any large-scale item selection/ranking domain;
- Auction/bidding mechanisms, prediction calibration, delivery optimization, or explore/exploit strategies (bandit algorithms, cold start);
- Model serving optimization, embedding-based retrieval, or vector databases;
- Columnar/OLAP databases (ClickHouse, BigQuery). AI/LLM tools for productivity.
Working style:
- High ownership - drives improvements end-to-end;
- Experiment-driven - benchmarks before committing, iterates based on evidence;
- Hands-on - reads code, debugs pipelines, digs into data.
