BuildSpark AI

Concepts

The practical map and cards for builders who already use AI.

Fourteen cards are live. The map lists the rest of the curriculum. Coming soon names are not links.

Start with Tokens

The unit models read, write, and bill in.

What are you trying to do?

Cut costPrompt Caching Improve answer qualityHallucination Pick a modelModel Selection & Routing Ask for a shaped replyStructured Outputs Ground answers in my docsRetrieval & Grounding Build an agentAgents Ship to productionShip & Maintain

What's new

Changelog
  • Agents track starter cards

    Agent loop, Tool calling, and Prompt injection are live cards. They cite OpenAI, Anthropic, and Gemini docs. Prompt injection also cites the OWASP Top 10 for LLM Applications. No rate or usage count was added.

    Agent loop, Tool calling, Prompt injection

  • Retrieval & Grounding starter cards

    RAG, Chunking, and Embeddings for retrieval are live cards. Foundations Embeddings stays without a card. The retrieval card links to that entry.

    RAG, Chunking, Embeddings for retrieval

  • Temperature and the Foundations pack

    Temperature is a live card. The Foundations pack page has a token-budget example and a pre-launch cost checklist. The example prices are labelled and cited. The traffic numbers on that page are labelled assumptions.

    Temperature

  • Four Foundations cards

    Prompt Caching, Model Selection & Routing, Structured Outputs, and Hallucination are live cards. Those four cite vendor API docs. The earlier cards still have no sources.

    Prompt Caching, Model Selection & Routing, Structured Outputs, Hallucination

  • Tokens and Context Window cost notes

    Cost on Tokens covers output versus input, cached input, and a higher rate past a size threshold. Cost on Context Window notes that resending the same history or documents adds up, and that prompt caching can make the repeated part cheaper.

    Tokens, Context Window

  • Card template v2

    Live cards use the v2 layout. Prerequisites and related concepts come from the registry.

    Tokens, Context Window, Prompting

  • Concepts site at the root

    The concepts product is the site root.

Next for you

Full map

Start with this card.

FoundationsTokens

Cards

Foundations Tokens

A token is a piece of text the model reads or writes. Limits and bills count tokens.

 Foundations Context Window

One request holds a fixed amount: instructions, history, documents, and the answer.

 Foundations Prompting

The same task changes when you name the reader, the shape, and what to leave out.

 Foundations Prompt Caching

Reusing an unchanged prefix so a repeated prompt can cost less than sending it from scratch.

 Foundations Model Selection & Routing

Picking a model, and sometimes a route between models, for the job in front of you.

 Foundations Structured Outputs

Asking for a reply that follows a shape, such as fields you name, so code can read it.

 Foundations Hallucination

A reply that states something the prompt and the sources do not support.

 Foundations Temperature

A sampling control that shifts how often less likely next tokens are picked.

 Retrieval & Grounding RAG

Retrieve relevant text, then ask the model to answer from that text.

 Retrieval & Grounding Chunking

Splitting documents into pieces that retrieval can store and return.

 Retrieval & Grounding Embeddings for retrieval

Using embeddings to find stored chunks that are close to a query.

 Agents Agent loop

A cycle of model step, tool result, and another model step, until a stop.

 Agents Tool calling

The model asks your code to run a named function, then reads the result.

 Agents Prompt injection

Untrusted text that tries to override the instruction or the tool rules.

Map

Four tracks. Live cards are links. The other names are not cards yet. Each track has a page written from the registry: Foundations, Retrieval & Grounding, Agents, Ship & Maintain.

Foundations

Tokens Context Window Prompting Prompt Caching Model Selection & Routing Structured Outputs Hallucination Temperature

Retrieval & Grounding

RAG Chunking Embeddings for retrieval

Agents

Agent loop Tool calling Prompt injection

Ship & Maintain

Evaluation Coming soon

Open the map

Ask

Keyword match against titles, summaries, and the text of live cards. Live cards come first. Other names are labelled coming soon. No model is called.

Apply packs

Concept cards are free. The Foundations pack includes a token budget set: one worked cost example and a pre-launch checklist. The other sets, and the other track packs, are not written yet.

Foundations pack All packs