好列表。

补充一些：

• 数据集工程 - https://huyenchip.com/machine-learning-systems-design/toc.html…
• 产品评估 - https://developers.openai.com/api/docs/guides/evaluation-best-practices…
• OpenAI Evals - https://developers.openai.com/api/docs/guides/evals…
• 上下文工程 - https://anthropic.com/engineering/effective-context-engineering-for-ai-agents…
• 智能体记忆与上下文生命周期 - https://platform.claude.com/cookbook/tool-use-context-engineering/context-engineering-tools…
• MCP 与工具生态 - https://modelcontextprotocol.io
• 可观测性与追踪 - https://docs.langfuse.com
• AI 智能体评估 - https://deepeval.com/guides/guides-ai-agent-evaluation…
• 推理优化（KV Cache、PagedAttention、Batching）- https://docs.vllm.ai
• 安全工程（Prompt 注入、智能体安全）- https://owasp.org/www-project-top-10-for-large-language-model-applications/…
• 业务指标与产品思维 - https://lennysnewsletter.com

令人惊讶的是，很多 AI 工程师花在调试检索、上下文、缓存、队列、权限和分析上的时间，比写 prompt 还要多。

机器学习系统设计

来源：https://huyenchip.com/machine-learning-systems-design/toc.html

机器学习面试

Chip Huyen huyenchip.com (https://huyenchip.com/) @chipro (https://twitter.com/chipro)

目录

1. 引言 (https://huyenchip.com/machine-learning-systems-design/research-vs-production.html#introduction-qzZkHeP)1. 研究 vs 生产 (https://huyenchip.com/machine-learning-systems-design/research-vs-production.html#research-vs-production-8DpYHKz)1. 性能要求 (https://huyenchip.com/machine-learning-systems-design/research-vs-production.html#performance-requirements-5DWpHYz) 2. 计算要求 (https://huyenchip.com/machine-learning-systems-design/research-vs-production.html#compute-requirements-eR1mHdR)
2. 设计机器学习系统 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#design-a-machine-learning-system-dwGQI5R)1. 项目设置 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#project-setup-zlkQIG9) 2. 数据管道 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#data-pipeline-698WI4R) 3. 建模 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#modeling-o97BIGR)1. 模型选择 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#model-selection-eRQEIDR) 2. 训练 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#training-5RLqIW9)1. 调试 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#debugging-89pbIkl) 2. 超参数调优 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#hyperparameter-tuning-BlADIyw) 3. 扩缩 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#scaling-49BpIQl) 4. 服务 (https://huyenchip.com/machine-learning-systems-design/design-a-machine-learning-system.html#serving-091rIYw)
3. 案例研究 (https://huyenchip.com/machine-learning-systems-design/case-studies.html#case-studies-bYrWS80)
4. 练习 (https://huyenchip.com/machine-learning-systems-design/exercises.html#exercises-rWl8SQW)

上一篇：引言 (https://huyenchip.com/machine-learning-systems-design/research-vs-production.html)

Mohit Goyal (Harness arc) (@ByteMohit)： 作为 AI 工程师，请学习：

>工程驾驭，而非仅仅 prompt 工程

>上下文工程，而非仅仅长 prompt

>Prompt 缓存 vs 语义缓存的权衡

>KV 缓存管理、驱逐、复用及大规模下的内存压力

>预填充 vs 解码延迟及