Job Description

ML Engineer — Owns the classical retail ML cores that power ARIP's most commercially impactful agents: A-22 Demand Forecasting (MAPE ≥5pts better than Oracle Demantra), A-23 Replenishment, A-24 Allocation, A-25 Causal Insight (DoWhy/EconML), and Wave 4 Commerce models (A-18 Pricing, A-19 Promotion, A-20 Markdown, A-21 Assortment). Consumes DEAA's MLOps platform (Model Registry, Drift Detection, Feature Store, Lotus's-curated wrappers) — does not rebuild OSS.

Remote candidates outside of Thailand are welcome to apply.

Key Responsibilities:

• Build, train, evaluate, and deploy classical retail ML models for W3 agents (A-22..A-25) and W4 Commerce model (A-18..A-21) using DEAA's OSS wrappers (Prophet / DoWhy / EconML / LightFM / scikit-learn / XGBoost)
• Register all models in DEAA's Model Registry (MLflow / Databricks Unity Catalog Models) with model cards; configure drift detection and feature store integration for every governed model
• Run A-22 side-by-side with Oracle Demantra during shadow phase; document Trust Gate progression evidence; prove MAPE ≥5 pts improvement by Wave 3 close
• Build A-25 Causal Insight models (DoWhy / EconML) with mandatory citation patterns and immutable Decision Log; A-25 is read-only — no transactional writes
• Partner with ARIP AI Engineers on model-to-agent integration contracts, latency budgets, fallback behaviour, and HITL gate criteria for ML-heavy agents
• Partner with Suite POs (Songkiat S3, Yaowaluck S1) on BU adoption, gate criteria, and per-model business value documentation

Requirements

• 5+ years building production ML systems in retail / commercial-decisioning: forecasting, replenishment, pricing, or recommendation at multi-store/multi-SKU scale
• Expert in retail forecasting (Prophet / statsmodels / NeuralProphet / DeepAR) and classical ML (XGBoost / LightGBM / scikit-learn)
• Causal inference in production (DoWhy / EconML) — required for A-25 Causal Insight Agent
• MLOps consumer: MLflow / Databricks Unity Catalog Models, Feature Store, drift detection — registers and operates governed models, not builds the platform
• Strong Python + Spark/PySpark; SQL fluency; Databricks (or equivalent lakehouse) production experience
• Calibre: Senior ML Engineer from Agoda (forecasting/pricing), Grab, Shopee, Lazada, Lotus's Australia / Tesco with retail ML production track record