机器学习工程师 I - 消息安全产品
Machine Learning Engineer I - Message Security Products
关于该职位
Abnormal AI 正在寻找一名机器学习工程师 - I（MLE）加入错误邮件检测（MED）团队。MED 团队在防止意外数据丢失方面发挥关键作用，通过检测和阻止错误发送的出站邮件，提供大规模保护，而不会给客户的 SOC 增加操作负担。
这是一个高度应用性的 MLE 职位，适合那些热衷于构建、迭代和实验的人。除了专注于模型训练外，你还将负责开发实用的端到端 ML 解决方案。这包括但不限于生成和优化特征、测试假设、平均信号，并将研究想法转化为生产级系统，同时跨职能协作，将客户需求转化为可衡量的产品改进。理想的候选人兼具动手实践的心态和技术严谨性，平衡创新与生产卓越，以推动实验、扩展解决方案并交付可靠的检测能力，在真实环境中产生有意义的客户影响。
你会做什么
- 与产品经理、技术负责人和工程相关方合作，确保技术交付符合路线图里程碑，并在支持的环境中成功发布通用版本（GA）。
- 负责错误邮件的完整 ML 生命周期，包括数据处理、特征工程、模型训练和评估、部署以及监控。持续交付可衡量的可靠性和客户影响的改进。
- 运行严格的实验和评估（离线指标、在线 A/B 测试、发布后监控），设置阈值，并进行有针对性的错误分析，以防止性能下降。
- 在不同时区之间有效沟通，保持高质量的技术文档，并为团队知识共享做出贡献。
- 参与所负责组件的轮班值班，职责集中在检测效果和实时评分系统上。优先事项包括解决与效果相关的警报、调查高可见度的误报，以及处理客户或内部团队报告的误报/漏报。
必备条件
- 计算机科学、机器学习、人工智能、信息系统或相关工程或量化领域的学士学位。
- 1 年以上在生产系统中构建和运营应用 ML 特征的经验。
- 有参与端到端 ML 系统开发的实证经验。
查看英文原文
About the Role
Abnormal AI is seeking a Machine Learning Engineer - I (MLE) to join the Misdirected Email Detection (MED) team. The MED team plays a critical role in preventing accidental data loss by detecting and blocking misdirected outbound emails, delivering protection at scale without adding operational burden to customer SOCs.
This is a highly applied role for MLEs who thrive on building, iterating, and experimenting. Rather than focusing solely on model training, you will also be responsible for developing practical, end-to-end ML solutions. This includes but is not limited to generating and refining features, testing hypotheses, averaging signals, and translating research ideas into production-grade systems, all while collaborating cross-functionally to turn customer needs into measurable product improvements. The ideal candidate combines a tinkerer's mindset with technical rigor, balancing innovation with production excellence to drive experimentation, scale solutions, and deliver reliable detection capabilities that create meaningful customer impact in real-world environments.
What you will do
- Partner with Product Manager, Tech Lead and engineering stakeholders to align technical deliverables to roadmap milestones and ensure successful GA launches across supported environments.
- Own the full ML lifecycle for Misdirected Email, including data wrangling, feature engineering, model training and evaluation, deployment, and monitoring. Deliver iterative improvements with measurable reliability and customer impact.
- Run rigorous experiments and evaluations (offline metrics, online A/B testing, post-launch monitoring), set thresholds, and conduct targeted error analysis to prevent regressions.
- Communicate effectively across time zones, maintain high-quality technical documentation, and contribute to shared team knowledge.
- Participate in shared on-call rotation for owned components, with responsibilities focused on detection efficacy and realtime scoring systems. Priorities include resolving efficacy-related alerts, investigating high-visibility false positives, and addressing reported false positives/false negatives from customers or internal teams.
Must Haves
- BS degree in Computer Science, Machine Learning, Artificial Intelligence, Information Systems, or a related engineering or quantitative field.
- 1+ years building and operating applied ML features in production systems.
- Proven experience contributing to end-to-end ML systems, including data wrangling (text and structured), feature engineering, model selection, training, evaluation, and production deployment with monitoring.
- Demonstrated ability to implement and reason about algorithms, develop features, average and combine signals, and apply numerical computing effectively.
- Demonstrated ability to interrogate production data, identify behavioral or trend shifts, and launch targeted experiments to improve model efficacy.
- Understanding of online vs offline pipelines, data tables and labeling workflows to effectively leverage tooling to support safe, scalable model deployments.
- Experience running offline metrics, online A/B tests, setting thresholds, and monitoring drift and performance, with guardrails and rollback strategies to ensure reliable iteration.
- Strong written and asynchronous communication skills. Effective working independently and across distributed, cross-functional teams.
Nice to Have
- Experience with our stack: Python, Go, AWS, Spark, Databricks
- Experience in email security/DLP or misdirected email prevention domains and customer-focused ML deployments.
- Experience writing detectors/rules to complement ML models for safe launches and rapid iteration.
- Experience with operationalising research into reliable, customer-facing systems, with emphasis on scalability, performance, and detection accuracy in real-world environments.
- Prior experience contributing to a small team or project to deliver a feature or component from scratch.
#LI-UC1
A note on AI in our process:
Abnormal AI uses AI-assisted tools to help our recruiting team prepare for candidate interviews. These tools analyze resume content and role requirements to suggest interview questions and areas for the interviewer to explore.They do not make hiring decisions or screen candidates automatically. Every decision about a candidacy is made by a person. Further, if your application is successful and Abnormal AI makes a conditional offer of employment, we will carry out pre-employment checks which must be successfully completed to progress to a final offer. All processes and pre-employment checks are in line with prevailing legislation and Abnormal AI's policies relevant to our security and privacy standards.
Abnormal AI is an equal opportunity employer. Qualified applicants will receive consideration for employment without regard to race, color, religion, sex, national origin, disability, protected veteran status or other characteristics protected by law. For our EEO policy statement please click here. If you would like more information on your EEO rights under the law, please click here.