机器学习工程师 – 威胁检测模型
ML Engineer – Threat Detection Models
职位地点：印度，远程优先
工作类型：全职
入职时间：2026年11月
经验要求：4年以上
语言要求：英语流利
行业：网络安全 / 企业SaaS / AI安全
关于这个职位
Pragmatike正在为一家全球企业网络安全公司招聘，该公司正在打造新一代产品，以保护AI代理、基于LLM的应用程序以及它们访问的数据。
威胁检测团队构建模型，实时判断提示、响应、工具调用或数据是否构成安全风险，包括提示注入、越狱尝试、敏感数据泄露、策略违规和异常代理行为。
你将负责威胁检测模型的全流程，从数据和训练到评估和生产部署，模型运行在基于Go的网关和端点服务中。
我们寻找一位ML工程师，既关注模型质量，也关注延迟、可靠性和生产性能，并且能够熟练使用Python和Go。
你将负责
- 使用分类器、基于嵌入的模型、微调的LLM以及规则/ML混合方法设计、训练和评估威胁检测器。
- 构建和维护训练和评估数据集、标注工作流和基准套件。
- 通过精确率/召回率、错误分析、漂移和对抗鲁棒性跟踪模型性能。
- 在严格的延迟要求下将模型部署到生产环境，使用Go构建推理服务并集成到网关和端点流程中。
- 通过量化、蒸馏、批处理和缓存等技术优化推理性能和成本。
- 与安全研究人员紧密合作，将新兴攻击技术转化为训练数据和检测逻辑。
- 构建和维护覆盖可复现训练、模型注册表、监控和安全模型发布的MLOps流程。
- 使用AI辅助开发流程加速实现、测试、调试和实验。
我们寻找的人选
- 4年以上在生产环境中构建和部署ML模型的经验，包括NLP或基于LLM的分类。
- 熟练的Python技能和现代ML栈经验，包括PyTorch、Hugging Face Transformers和scikit-learn。
- 有微调基于Transformer的模型的实际经验。
- 扎实的Go技能，或有后端工程经验，能够快速掌握Go。
查看英文原文
Location: India, Remote-First
Employment: Full-Time
Start Date: November 2026
Experience: 4+ years
Language: Fluent English required
Industry: Cybersecurity / Enterprise SaaS / AI Security
ABOUT THE OPPORTUNITY
Pragmatike is recruiting on behalf of a global enterprise cybersecurity company building a new generation of products to secure AI agents, LLM-powered applications, and the data they access.
The Threat Detection team builds the models that determine, in real time, whether a prompt, response, tool call, or piece of data presents a security risk, including prompt injection, jailbreak attempts, sensitive data exposure, policy violations, and anomalous agent behavior.
You’ll own detection models end to end, from data and training through evaluation and production serving, with models running inline in Go-based gateway and endpoint services.
We’re looking for an ML Engineer who cares as much about latency, reliability, and production performance as model quality, and who is comfortable working across both Python and Go.
WHAT YOU’LL DO
- Design, train, and evaluate threat detectors using classifiers, embedding-based models, fine-tuned LLMs, and rule/ML hybrid approaches.
- Build and maintain training and evaluation datasets, labeling workflows, and benchmark suites.
- Track model performance through precision/recall, error analysis, drift, and adversarial robustness.
- Serve models in production under strict latency requirements, working in Go to build inference services and integrate with gateway and endpoint pipelines.
- Optimize inference for performance and cost through techniques such as quantization, distillation, batching, and caching.
- Work closely with security researchers to turn emerging attack techniques into training data and detection logic.
- Build and maintain MLOps workflows covering reproducible training, model registries, monitoring, and safe model rollouts.
- Use AI-assisted development workflows to accelerate implementation, testing, debugging, and experimentation.
WHAT WE’RE LOOKING FOR
- 4+ years of experience building and deploying ML models in production, including NLP or LLM-based classification.
- Strong Python skills and experience with the modern ML stack, including PyTorch, Hugging Face Transformers, and scikit-learn.
- Hands-on experience fine-tuning transformer-based models.
- Solid Go skills, or strong backend engineering experience with the ability to become productive in Go quickly.
- Experience with low-latency model serving, using technologies such as ONNX Runtime, TorchServe, Triton, vLLM, or custom serving infrastructure.
- Strong evaluation discipline, including dataset design, metrics, error analysis, and adversarial testing.
- Experience with Docker, Kubernetes, and at least one major cloud provider.
- Fluent English with strong written and verbal communication skills.
- Comfortable using modern AI coding assistants such as Claude Code, Cursor, GitHub Copilot, Codex, or similar. This is a must-have.
- Strong ownership and the ability to work independently in a remote-first, distributed environment.
NICE TO HAVE
- Experience building detection systems for prompt injection, jailbreaks, or content safety.
- Experience with DLP or sensitive-data classification, including PII, secrets, or source code.
- Familiarity with adversarial ML and model robustness techniques.
- Previous experience in cybersecurity, security tooling, or trust & safety.
- Experience introducing AI-assisted or agentic development workflows across engineering teams.
AI-FIRST ENGINEERING
AI is a core part of the engineering workflow on this team.
During the interview process, you'll be asked about how you use AI in real-world software development, including the tools you use, how you validate their output, and where AI has changed the way you work.
We're looking for engineers who use AI as a force multiplier for quality, productivity, and problem-solving, not simply as an autocomplete tool.
WHY JOIN
- Build real-time AI threat detection systems at the intersection of cybersecurity and agentic AI.
- Take models from research and experimentation all the way into low-latency production environments.
- Work on challenging problems spanning ML, LLM security, adversarial AI, and distributed systems.
- Own a critical detection workstream and influence how AI security threats are identified and mitigated.
- Collaborate with a highly technical, distributed team where AI is a core part of the development process.
Pragmatike is committed to a fair, transparent, and inclusive recruitment process. We do not discriminate based on age, disability, gender, gender identity or expression, marital or civil partner status, pregnancy or maternity, race, religion or belief, sex, or sexual orientation.
In accordance with GDPR, your personal data will be processed lawfully, fairly, and securely and used solely for recruitment purposes, including sharing it with our client(s) for employment consideration.