技能 senior-ml-engineer
🤖

senior-ml-engineer

安全 🌐 网络访问⚙️ 外部命令

Deploy Production ML Models

也可从以下获取: davila7

Building ML pipelines requires expertise in model deployment, monitoring, and MLOps practices. This skill provides production-ready tools for deploying ML models, integrating LLMs, and building scalable RAG systems with enterprise-grade reliability.

支持: Claude Codex Code(CC)
🥈 77 白银
1

下载技能 ZIP

2

在 Claude 中上传

前往 设置 → 功能 → 技能 → 上传技能

3

开启并开始使用

测试它

正在使用“senior-ml-engineer”。 Help me deploy my trained TensorFlow model for real-time inference with sub-50ms latency

预期结果:

  • Containerize the model using Docker with TensorFlow Serving
  • Create Kubernetes deployment with HPA for auto-scaling
  • Implement batch inference for throughput optimization
  • Add Prometheus metrics for latency, throughput, and error rates
  • Configure load balancer with health checks and circuit breakers

正在使用“senior-ml-engineer”。 Build a RAG system that queries my internal knowledge base with citation support

预期结果:

  • Set up embedding model and vector database (Pinecone or Weaviate)
  • Implement document chunking with overlap for context preservation
  • Create hybrid search combining semantic and keyword matching
  • Add source citation and confidence scoring to responses
  • Implement re-ranking for improved result quality

安全审计

安全
v3 • 1/16/2026

All 78 static findings evaluated as false positives. The static scanner incorrectly flagged documentation text (headings, bullet points) containing common English words as 'weak cryptographic algorithm' patterns. Python scripts contain only standard library imports and logging. Bash code blocks in SKILL.md are documentation examples, not shell execution. No malicious code exists in this skill.

9
已扫描文件
1,614
分析行数
2
发现项
3
审计总数
审计者: claude 查看审计历史 →

质量评分

68
架构
100
可维护性
87
内容
23
社区
100
安全
91
规范符合性

你能构建什么

Deploy models to production

Package and deploy trained ML models with Docker, Kubernetes, and REST API endpoints

Build RAG applications

Create retrieval-augmented generation systems using LlamaIndex and LangChain

Implement MLOps

Set up model versioning, monitoring, and automated retraining pipelines with MLflow

试试这些提示

Deploy model
Help me deploy my trained PyTorch classification model to production. I need a REST API endpoint, Docker container, and Kubernetes deployment configuration. Include monitoring for model latency and accuracy drift.
Build RAG system
Build a RAG system for my documentation. I have PDF and markdown files. Use a vector database, implement chunking strategies, and create a query interface that works with my internal LLM API.
MLOps setup
Set up MLOps for my machine learning team. We need MLflow for model versioning, automated training pipelines triggered by data changes, and alerts for model performance degradation.
Advanced LLM integration
Design an enterprise LLM integration architecture. Include prompt caching, rate limiting, cost tracking by team, and integration with our existing authentication system. Support both OpenAI and self-hosted models.

最佳实践

  • Design for 10x scale from day one with horizontal scaling architecture
  • Implement comprehensive monitoring including latency, drift detection, and data quality checks
  • Use feature stores to ensure consistency between training and inference

避免

  • Deploying models without version control or rollback capabilities
  • Skipping model drift detection until production issues arise
  • Hardcoding API keys or credentials instead of using secrets management

常见问题

What ML frameworks does this skill support?
PyTorch, TensorFlow, Scikit-learn, and XGBoost are fully supported with deployment tools.
Can I use this for LLM fine-tuning?
The skill focuses on deployment and integration. For fine-tuning, use Claude's codex capabilities.
Does this include cloud credentials?
No. You provide your own AWS, GCP, or Azure credentials for deployment.
What monitoring tools are integrated?
MLflow, Weights & Biases, and Prometheus for metrics, plus custom dashboard templates.
Can I deploy on-premises?
Yes. Docker and Kubernetes configurations work on any K8s cluster or container runtime.
How do I handle model updates?
Use feature flags and canary deployments for safe rollouts with quick rollback capability.