AI Cheat Sheets

One-page AI cheat sheets for developers: how each mechanism works, the rules that hold, and when to use it. Click a sheet to watch it build step by step.

  • Home /
  • AI Cheat Sheets

Embedding cheat sheet

Distance carries meaning only inside one model's space, so the index and every query must share that model.

#Retrieval-Augmented GenerationAnimated

→ Topic: Embedding

Encoder-Decoder Architecture cheat sheet

One half reads the whole input, the other writes the output token by token, and cross-attention lets the writer look back at every input position.

#Transformer & Attention InternalsAnimated

→ Topic: Encoder-Decoder Architecture

Eval harness: reading AI scores cheat sheet

An eval harness keeps the questions, the prompt and the grading fixed so scores compare. Follow one real question through it and see one model score 0 % to 63 %.

#Model Evaluation

→ Read the article

Evaluation Harness cheat sheet

Hold the tests and the scoring fixed, swap only the model, and the scores become comparable.

#Model EvaluationAnimated

→ Topic: Evaluation Harness

Fine-Tuning cheat sheet

Fine-tuning shapes how a model responds; new facts belong in retrieval, not in the weights.

#LLM Training & Pre-TrainingAnimated

→ Topic: Fine-Tuning

Generative Adversarial Network cheat sheet

The generator improves only as long as the discriminator keeps sending a useful signal, so balance decides whether training works.

#Neural Network ArchitecturesAnimated

→ Topic: Generative Adversarial Network

Generative Media APIs cheat sheet

Long media jobs break a held-open call; the webhook path needs duplicate-safe handling from the start.

#Generative Media Pipelines & APIsAnimated

→ Topic: Generative Media APIs

Generative Media Pipelines cheat sheet

One generation call is the easy part; the queue, the gate and an idempotent publish step decide what safely ships.

#Generative Media Pipelines & APIsAnimated

→ Topic: Generative Media Pipelines

Graph Neural Network cheat sheet

Each round lets a node learn from its neighbors; keep the edges explicit and cap depth before node signals blur together.

#Neural Network ArchitecturesAnimated

→ Topic: Graph Neural Network

Hallucination cheat sheet

A language model predicts what sounds true; grounding and claim checks shrink the gap but never close it.

#Safety & Red TeamingAnimated

→ Topic: Hallucination

Human Evaluation for AI cheat sheet

A score counts only after raters agree on the rubric; low agreement sends you back to rewrite it, not to hire raters.

#Model EvaluationAnimated

→ Topic: Human Evaluation for AI

Human-in-the-Loop for Agents cheat sheet

Gate only the steps you cannot undo, keep the pause resumable, and plan what happens when nobody answers.

#Agent Reliability & OperationsAnimated

→ Topic: Human-in-the-Loop for Agents

Image Upscaling cheat sheet

Interpolation only blends pixels it already has; AI super-resolution invents plausible detail, so treat the output as a guess.

#AI Image Generation & EditingAnimated

→ Topic: Image Upscaling

Inference cheat sheet

Each token waits on the one before it, so more GPUs serve more requests rather than one response faster.

#Inference OptimizationAnimated

→ Topic: Inference

Knowledge Graphs for RAG cheat sheet

A graph adds the links vector search cannot follow; extend the vector index rather than replacing it.

#Knowledge Retrieval SystemsAnimated

→ Topic: Knowledge Graphs for RAG

LLM Cost Management cheat sheet

Most token spend is structural waste, so caching, tiering and batching cut the bill without making answers worse.

#LLMOps & ProductionAnimated

→ Topic: LLM Cost Management

LLM Fallback and Retry Patterns cheat sheet

Transient errors earn a retry with jittered backoff, permanent ones never do, and a fallback model takes over once retries run out.

#LLMOps & ProductionAnimated

→ Topic: LLM Fallback and Retry Patterns

LLM Gateway cheat sheet

Centralizing every model call buys control, and then the gateway itself needs a fallback path.

#LLMOps & ProductionAnimated

→ Topic: LLM Gateway

LLM Load Testing cheat sheet

Under concurrent load, the first token waits for a KV-cache slot, so the p99 tail breaks long before the average moves.

#LLMOps & ProductionAnimated

→ Topic: LLM Load Testing

LLM Logging and Auditing cheat sheet

Every model call becomes a structured record; personal data is masked before it reaches storage, and an audit reviews what was kept.

#LLMOps & ProductionAnimated

→ Topic: LLM Logging and Auditing

LLM Observability cheat sheet

A trace shows which step failed and why; a status code or the final output alone never does.

#LLMOps & ProductionAnimated

→ Topic: LLM Observability

LLM orchestration: do you need it? cheat sheet

LLM orchestration coordinates the steps of an AI workflow, including decisions and recovery when a process fails. See the shapes and their limits.

#AI Agent Architecture

→ Read the article

LLM-as-a-Judge cheat sheet

A judge model scales grading, but its verdicts earn trust only after a precise rubric is checked against human labels.

#Model EvaluationAnimated

→ Topic: LLM-as-a-Judge

Long-Context vs RAG cheat sheet

Long context lets attention sort the whole corpus; RAG filters before the prompt, so each fails in its own way.

#Retrieval-Augmented GenerationAnimated

→ Topic: Long-Context vs RAG

LoRA for Image Generation cheat sheet

A LoRA is a small file of real learned weights, shaped for one base model and invoked by its trigger word.

#AI Image Generation & EditingAnimated

→ Topic: LoRA for Image Generation

LoRA vs QLoRA vs full fine-tuning cheat sheet

LoRA fine tuning trains small added weights; QLoRA also reduces base-model memory. Compare both with full fine-tuning and choose a method.

#LLM Training & Pre-Training

→ Read the article

MCP architecture cheat sheet

MCP architecture connects an AI host to a server through a client that exchanges messages to use tools, resources and prompts.

#Agentic & Autonomous Coding

→ Read the article

Metadata Filtering cheat sheet

Where the filter runs, before, during or after the search, decides both recall and speed.

#Knowledge Retrieval SystemsAnimated

→ Topic: Metadata Filtering

Mixture of Experts cheat sheet

Each token wakes only a few experts, so compute follows active parameters while the router has to keep the load balanced.

#Transformer & Attention InternalsAnimated

→ Topic: Mixture of Experts