LLM engineering
- Category
- AI and LLMs
- Docs outline
- 14 sections
Official docs outline
platform.claude.com- Working with messages
Roles, multi-turn conversation state, system prompts, and the stateless request model of LLM APIs.
- Context windows and tokens
Context limits, tokenization, token counting, and managing long conversations.
- Prompt engineering
Clear instructions, examples, structure, and role prompting to improve output quality.
- Structured outputs
Getting schema-conforming JSON from models for reliable parsing.
- Tool use and function calling
Defining tools, handling tool calls, and agent loops that act on external systems.
- Streaming
Streaming tokens incrementally to cut perceived latency.
- Extended thinking and reasoning
Trading tokens for deeper reasoning with thinking budgets.
- Embeddings and retrieval
Vector embeddings, semantic search, and the basis of retrieval-augmented generation.
- Citations and grounding
Grounding answers in source documents and reducing hallucinations.
- Prompt caching
Reusing long prompt prefixes to reduce cost and latency.
- Evaluation and testing
Defining success criteria and building test sets to measure LLM application quality.
- Guardrails and safety
Mitigating jailbreaks and prompt injection and increasing output consistency.
- Batch processing
Asynchronous high-volume inference at lower cost.
- Rate limits and errors
Token and request quotas, retry strategy, and error handling in production.
Each section links to the official page. Titles and summaries by Extra Credit; the docs belong to their authors.
