全职 Data Analyst Engineer 工作, 薪水, Beyondsoft Malaysia Federal Territory 公司招聘中 - Ricebowl

Job Description

We are hiring a core partner to co-own our next-gen anomaly detection & attribution (ADA)

platform that powers real-time monitoring, automated diagnosis, and decision support across

multiple games/regions. You will lead end-to-end work—architecture, development,

maintenance, and analytical impact—while advancing our LLM-Agent capabilities to turn

noisy signals into clear, actionable narratives for business teams.

What You’ll Do

Own the ADA platform lifecycle: design, implement, and maintain robust pipelines for T+1/near-real-time anomaly detection, multi-baseline benchmarking (global/region/country), and multi-source attribution (holidays, versions, events, migrations, user behavior).

Advance detection & inference: productionize change-point/outlier methods, time-series

features, causal/ablation checks, and automated “storyline” generation that explains what happened, why, and what to do next.

Build AI Agents around the system: design tool-use and reasoning flows (ReAct/LangGraph or similar) to enable conversational drill-downs for ops/PMs.

Productize insights: ship dashboards/alerts (email/Chat/WeCom) and concise decision memos; iterate with stakeholders in publishing, ops, marketing, and analytics.

Collaboration: partner with DS/Eng/PM to scope, roadmap, and ship; document architecture, APIs, and runbooks.

Job Requirements

Must-Have Qualifications

Technical core:

Strong Python engineering (clean code, testing, packaging); solid SQL for large analytical workloads.

Hands-on with time-series/anomaly detection (change-point, robust stats,

seasonality/holiday adjustment, multivariate signals) and attribution logic.

Practical exposure to LLM application patterns (tool calling, function calling, retrieval/RAG, Agent planning), and at least one framework/API (OpenAI/Claude/DeepSeek, LangChain/LangGraph, etc.).

Systems/product mindset: ability to translate business pain points into measurable

detection/attribution logic and ship reliable features on short cycles.

Ownership & reliability: you build guardrails, monitors, and docs; you debug in production and prevent regressions.

Nice-to-Have / Preferred

Model eval & prompt engineering: rubric design, offline eval sets, golden tasks, prompt/test versioning, data flywheels.

Causal & experimentation: diff-in-diff, CUPED, synthetic controls, online AB testing at scale.

Dashboards & alerts: Superset/Tableau/Looker/Metabase; alerting via Slack/WeCom/Email with noise-reduction heuristics.

Gaming analytics domain: retention funnels, reactivation, event/version rollout attribution,fraud/smurf detection.

Infra & ops: Docker/K8s, CI/CD, IaC; cloud stacks (GCP/AWS/Azure); cost/perf tuning