Fast Answers for AI Engineering Decisions

Use this section when you need the short answer before the deep dive. Each guide gives a practical default, the trade-offs, and a link to the longer article when the details matter.

The blog archive stays focused on long-form articles. Fast Answers are shorter decision guides.

AI agents

QuestionShort answerGuideDeep dive
Which agent framework should I start with?Use LangChain agents for a high-level start, LangGraph for explicit state, or a framework aligned with the product’s retrieval and platform needs.Best AI Agent FrameworksAI Agent Reasoning Loops
Which agent evaluation tool should I use?Use Phoenix for open tracing, LangSmith for a managed workflow, DeepEval for Python CI, or Promptfoo for local matrix and red-team tests.Best AI Agent Evaluation ToolsAgent Evals: Traces to Test Suites
How should I secure an AI agent?Remove power first: narrow tools, pre-tool policy checks, sandboxes, scoped credentials, approvals, egress controls, and traces.AI Agent Security ChecklistAI Agent Security
Which tool interface should I use?Use function calling inside one app, MCP for reusable servers, a CLI for mature commands, or skills for reusable workflow instructions.AI Agent Tool InterfacesAI Agent Tool Use

Model engineering and infrastructure

QuestionShort answerGuideDeep dive
Which local LLM tool should I use on macOS?Compare Ollama, LM Studio, llama.cpp, and MLX-LM by workflow, artifact, runtime control, and memory headroom.Local LLM Tools on macOSLocal LLMs on macOS
Which quantization representation should I use?Choose the runtime and hardware first; GGUF, AWQ, GPTQ, FP8, and NF4 describe different layers or goals.LLM Quantization FormatsModel Quantization Guide
Should I prompt, use RAG, or fine-tune?Prompt for instructions, use RAG for changing knowledge and evidence, and fine-tune for repeatable behavior.Fine-Tuning vs RAG vs PromptingLLM Fine-Tuning Guide

Retrieval and language systems

QuestionShort answerGuideDeep dive
Which RAG evaluation stack should I use?Start with deterministic stage metrics, then add calibrated judges and product-specific checks.Best RAG Evaluation ToolsRAG Evaluation Metrics
Which OCR model should I use for document AI?Use classical OCR for clean print, a specialized parser for structured pages, and a document VLM when visual reasoning is required.Best OCR ModelsOCR Guide
Which NER model should I use?Use spaCy for stable labels, GLiNER for changing labels, a supervised encoder when you have data, or an LLM for reasoning-heavy records.Best NER ModelsNER Guide
What search ranking stack should I build?Start with BM25, add dense recall, fuse complementary results, then rerank a latency-bounded shortlist.BM25 vs Embeddings vs RerankersSearch Ranking Stack

All fast answers 11

  1. Model Engineering · Retrieval and Language Systems

    Fine-Tuning vs RAG vs Prompting: Which Should You Use?

    Choose prompting, retrieval-augmented generation, or fine-tuning by whether the problem is instructions, changing knowledge, or model behavior.

    · 2 min