Core Concepts

Before building complex workflows, it's essential to understand the fundamental components of modern AI applications.

Large Language Models (LLMs)

The engine of the workflow. Models like GPT-4, Claude, or Llama process text (and images) to generate responses. Understanding their context window, latency, and reasoning capabilities is key to choosing the right one.

Prompts & Prompt Engineering

The interface to the model.

• Zero-shot: Asking without examples.
• Few-shot: Providing examples to guide the model's style and logic.
• Chain-of-Thought: Asking the model to "think step-by-step" to improve reasoning.

Context Management

LLMs have no memory of past requests unless you provide it. Managing context involves:

• Conversation History: Summarizing or truncating past turns.
• Context Window Optimization: Ensuring you don't exceed the model's limits.

Retrieval Augmented Generation (RAG)

Combining parametric memory (what the model knows) with non-parametric memory (your data).

1. Retrieve: Search a vector database or API for relevant information.
2. Augment: Insert that information into the prompt.
3. Generate: Ask the LLM to answer using the retrieved context.

Structured Output

For programmatic workflows, we often need JSON or XML, not free text. Techniques include:

• Function Calling: Defining tool schemas the LLM can "call".
• Json Mode: Constraining the model to valid JSON.