We are seeking a Senior Applied Scientist to join our team. In this role, you will build and operate production recommendation and personalization systems that serve millions of users across our retail and ecosystem platforms. The core expectation is machine learning excellence: recommendation models, learning-to-rank systems, personalization pipelines, feature engineering, offline/online evaluation, and production serving at scale.

MAIN RESPONSIBILITIES

• Own the end-to-end lifecycle of recommendation and personalization products, from problem definition, model design, and training to deployment, monitoring, and continuous improvement.
• Design, train, and deploy recommendation models across marketplace and retail surfaces, including homepage, search, category pages, cart, and cross-sell/upsell placements.
• Build and maintain the full recommendation stack: candidate generation, ranking, re-ranking, and business-rule layers.
• Apply foundational RecSys techniques in production, including collaborative filtering, content-based filtering, matrix factorization, and factorization machines.
• Develop modern recommendation approaches, including two-tower retrieval models, graph-based recommenders, and sequential/session-based models that capture user behavior over time.
• Model user-item-context relationships across categories, merchants, locations, price sensitivity, and lifecycle stage.
• Build learning-to-rank systems that optimize for business objectives such as conversion, gross merchandise value, retention, and long-term customer value, subject to relevance and fairness constraints.
• Design personalization features from transaction, behavioral, catalog, and contextual data, including user affinity, item popularity, freshness, price affinity, and cross-category relationships.
• Implement calibrated ranking models and monotonic constraints where business logic requires predictable behavior.
• Build segment-aware and lifecycle-aware personalization, including new-user cold-start, new-item cold-start, and re-engagement strategies.
• Work with product and business teams to translate commercial goals into model objectives, label definitions, and evaluation metrics.
• Build robust training and serving feature pipelines with point-in-time correctness to avoid data leakage.
• Design feature stores and offline/online feature definitions that remain consistent between training and inference.
• Handle missingness, categorical cardinality, temporal drift, and delayed labels appropriately in production pipelines.
• Partner with data engineering on batch and streaming data sources, including transactions, catalog changes, user events, and inventory state.
• Build comprehensive offline evaluation pipelines using standard RecSys metrics: NDCG, Recall@K, MRR, MAP, Hit Rate@K, and AUC.
• Design held-out and time-based validation that reflects production conditions, including cold-start users, new items, and temporal distribution shift.
• Run and analyze online experiments through A/B testing, including experiment design, power analysis, guardrail metrics, and decision readouts.
• Apply counterfactual and off-policy evaluation where direct A/B testing is limited.
• Define model acceptance criteria before launch and monitor performance after launch against those criteria.
• Build and maintain high-performance inference services (FastAPI or equivalent) with low latency, high availability, and graceful degradation.
• Implement candidate generation and ranking services that meet strict latency budgets, including caching, batching, and fallback strategies.
• Manage model versioning, reproducibility, rollback, and release processes.
• Monitor model health in production: prediction drift, feature drift, latency, error rates, coverage, and business metric impact.
• Work with Docker and Linux environments; collaborate with platform teams on deployment, autoscaling, and observability.

REQUIRED EXPERIENCE AND SKILLS

• Bachelor's degree in Computer Science, Data Science, or a related field.
• Strong software engineering foundation with a solid understanding of algorithms, data structures, and system design principles.
• Strong understanding of machine learning fundamentals: supervised learning, regularization, generalization, calibration, and evaluation methodology.
• Hands-on production experience building recommendation or ranking systems, not only research or coursework.
• Deep knowledge of Recommendation Systems (RecSys), spanning foundational techniques: Collaborative Filtering, Content-based Filtering, Matrix Factorization, and Factorization Machines.
• Hands-on experience with modern RecSys paradigms, including:

- Graph-based Recommenders for modeling user-item-context relationships.

- Sequential / Session-based Recommenders for capturing user behavior over time.

- Two-tower retrieval models and embedding-based candidate generation at scale.

• Practical experience with modern ranking models, such as gradient-boosted decision trees, deep learning rankers, or sequence models.
• Experience with retrieval + ranking architectures, including candidate generation, approximate nearest-neighbor search, and re-ranking layers.
• Experience with cold-start handling, exploration strategies, and popularity/freshness trade-offs.
• Strong command of offline RecSys evaluation metrics (NDCG, Recall@K, MRR, MAP, Hit Rate@K, AUC).
• Experience designing and interpreting A/B tests, including guardrail metrics and statistical significance.
• Understanding of position bias, exposure bias, feedback loops, and other common pitfalls in recommendation evaluation.
• Ability to define model acceptance criteria and communicate uncertainty honestly to technical and business stakeholders.
• Proficiency in Python (primary) and familiarity with TypeScript (secondary).
• Strong SQL skills and experience working with large behavioral and transaction datasets.
• Experience with ML frameworks such as PyTorch, TensorFlow, or XGBoost/LightGBM.
• Experience with feature stores, data pipelines, and point-in-time-correct feature computation.
• Hands-on experience with Docker and comfortable working in Linux environments.
• Familiarity with workflow orchestration and data processing tools such as Airflow, Spark, dbt, or equivalent.
• Experience with vector search or embedding retrieval systems for candidate generation.
• Experience deploying models to production and operating them after launch.
• Comfortable with API design, service monitoring, logging, and incident response for ML systems.
• Ability to collaborate with backend, data engineering, product, and business teams.
• Clear written and verbal communication, including the ability to explain model behavior and evaluation results to non-ML stakeholders.

NICE TO HAVE

• Experience with uplift modeling, causal inference, or counterfactual evaluation.
• Familiarity with multi-objective ranking and business-constrained optimization.
• Experience with real-time personalization using streaming event data (Kafka, Redis PubSub, or similar).
• Experience with MLOps tooling, model registries, and automated retraining pipelines.
• Working knowledge of LLM-based systems is a bonus, but not a core requirement for this role.
• Familiarity with LLM-augmented recommendation, conversational recommenders, or embedding-based retrieval using foundation models.
• Exposure to RAG systems, vector databases, or hybrid search.
• Interest in using LLMs to improve feature generation, metadata enrichment, or explanation of recommendation results.
• Experience with LLM evaluation, caching, or cost/latency optimization is a plus.
• Candidates do not need prior agent-framework experience to be considered. If you have strong recommendation, ranking, and personalization experience, we want to hear from you.

BENEFITS AND PERKS
Salary & Allowances

• 13-month salary with annual performance bonus, project incentives, sales incentives (based on position)
• Lunch allowance: 730.000 VND/month
• Special occasion bonus: 2.500.000VND/year
• Annual leaves: Up to 20 days/year (based on levels)
• Health: Social insurance, premium health insurance, yearly health check 
• Laptop, screen and other needed facilities/accounts/tools for work