Staff Engineer, AI/ML

RIO's promise is AI-native network operations. As a Senior AI Engineer on the AI Model & Harness team, you will span two equally important halves: developing the ML models that detect anomalies, predict faults and localize root causes across telecom networks; and building the AI harness - the training, evaluation, inference and monitoring infrastructure - that takes those models from notebook to production with confidence. You will work directly on top of RIO's ClickHouse and YugabyteDB (YBSQL) data platforms, in tight partnership with the Data Management Pipelines team.

Anomaly detection: Develop and productionize multivariate time-series anomaly detection models over network KPIs and counters to surface degraded cells and network elements before outages.

Predictive fault & incident analysis: Build forecasting and classification models that predict faults and incidents ahead of failure, and correlate alarms across domains for root-cause analysis and noise reduction.

Model portfolio: Own models across the operations lifecycle - KPI forecasting, event correlation, root cause localization, capacity prediction and optimization - from framing through production monitoring.

AI-defined KPIs: Move beyond vendor KPI formulas by learning leading/lagging indicators directly from raw counters, surfacing signals humans and rule-based systems miss.

Training & experimentation: Build reproducible training, experiment-tracking and hyperparameter-tuning workflows over large-scale telemetry datasets drawn from ClickHouse/YBSQL.

Evaluation frameworks: Design rigorous offline and online evaluation (backtesting, precision/recall on incident ground truth, A/B and shadow deployments) so every model's operational impact is measurable.

Serving & inference pipelines: Deploy low-latency, high-availability inference services with canary/rollback strategies; implement batch and streaming inference over near-real-time pipeline data.

Monitoring & lifecycle: Own model observability - drift detection, data-quality gates, alerting on degradation - and automated retraining/evaluation loops.

LLM & agentic AI: Contribute guard-railed LLM-based capabilities such as RAG over fault manuals and specifications, natural-language intent interfaces and LLM-assisted root-cause analysis.

Partnership with data pipelines: Define feature and training-data contracts with the Data Management Pipelines team; ensure models consume and produce data through governed, well-modeled interfaces.

Operational excellence: Participate in on-call for AI services; own the reliability of everything the harness runs in production.

Our worldwide practices describe specific behaviours that make Rakuten unique and united across the world. We expect Rakuten employees to model these 5 Shugi Principles of Success.

Always improve, always advance. Only be satisfied with complete success - Kaizen.

Be passionately professional. Take an uncompromising approach to your work and be determined to be the best.

Hypothesize - Practice - Validate - Shikumika. Use the Rakuten Cycle to success in unknown territory.

Maximize Customer Satisfaction. The greatest satisfaction for workers in a service industry is to see their customers smile.

Speed!! Speed!! Speed!! Always be conscious of time. Take charge, set clear goals, and engage your team.