Prompt recipe card
Use this five-part recipe for reliable prompts. Not every part is required — but skipping **Task** and **Format** is the most common mistake
Reference
Cheatsheet library
Decision cards and checklists — print or pin beside your desk.
27 cheatsheets
Prompt vs RAG vs fine-tune
Pick the smallest technique that solves the job. Upgrade only when the simpler option fails.
Safe sharing checklist
Run through this before pasting anything into a chatbot — especially work, school, or health info.
Verify AI answers checklist
Confident tone ≠ correct facts. Use this when answers matter for school, work, health, money, or law.
Pick a model for your task
Match the job to model type before chasing the biggest name on the leaderboard.
Token & cost quick reference
Rough math for planning prompts, RAG, and API budgets.
Prompt injection defense patterns
Layer defenses — no single filter catches everything.
RAG quality checklist
If answers are wrong, walk the pipeline **retrieval first**, generation second.
Streaming vs batch inference
- Interactive chat or copilot
Local vs cloud models
- Sensitive docs, offline field work, air-gapped labs
Agent loop cheatsheet
Think → Act (tool) → Observe → Think → …
Chat memory strategies
1. Keep system prompt + last N turns + summary
Literacy trust checklist
Use when deciding whether to **believe, share, or act on** AI-generated text, images, audio, or video.
Embedding dimensions cheat sheet
Vectors have a **fixed length (dimensions)** per model — index and queries must match.
Production RAG ship card
Image prompt card
Hybrid search patterns card
**Pipeline:** keyword top-20 + vector top-20 → dedupe → re-rank → top-5 to prompt
Agent ship checklist
Eval gate card
Guardrails layers card
LoRA vs RAG decision card
Research bot ship card
ML Python starter card
Minimum Python before AI/ML labs in Lane D.
Production ops signals
What to watch after shipping RAG or agents.
Canary ship card
Safe rollout checklist for models, prompts, and RAG indexes.
Re-rank patterns
Retrieve wide, re-rank narrow.
AI incident response card
- Trace IDs from failing requests