Building Systems with the ChatGPT API - DeepLearning.AI

Teaches the pattern of sequencing multiple API calls where outputs from prior completions feed as inputs to subsequent prompts, enabling complex reasoning workflows. The course demonstrates how to structure Python code that maintains context across multiple ChatGPT API calls, allowing each step to build on previous results without re-sending full conversation history each time.

Demonstrates using ChatGPT API to classify incoming user queries into predefined categories, then routing to appropriate downstream handlers or prompts based on classification results. The approach uses the LLM itself as a classifier rather than separate ML models, with the classification prompt designed to output structured category labels that code can parse and act upon.

Teaches how to maintain and manage conversation history in multi-turn interactions with ChatGPT API, including strategies for managing context window limits, summarizing long conversations, and deciding what information to retain or discard. The course demonstrates how to structure Python code that maintains conversation state and passes appropriate context to each API call.

Teaches how to use ChatGPT API to evaluate user inputs and system outputs for safety, policy violations, and harmful content. The approach involves crafting moderation prompts that ask the LLM to assess content against defined safety criteria and return structured judgments that can trigger filtering, flagging, or rejection logic.

Teaches prompting techniques where ChatGPT is instructed to break down complex problems into intermediate reasoning steps, with the ability to validate or evaluate each step before proceeding. The course demonstrates how to structure prompts that elicit step-by-step reasoning and how to parse and validate intermediate outputs to ensure correctness before using them in downstream logic.

Teaches using ChatGPT API to evaluate the quality, correctness, and relevance of LLM-generated outputs by crafting evaluation prompts that assess outputs against defined criteria. The approach involves using a second LLM call to judge the quality of a first LLM call, enabling automated quality gates and feedback loops without manual review.

Teaches how to craft system prompts that define the personality, constraints, and behavior of a ChatGPT-powered system, ensuring consistent responses across multiple user interactions. The course covers how system prompts interact with user messages and how to structure them to enforce specific behaviors, tone, and knowledge boundaries.

Teaches techniques for designing prompts that elicit structured, machine-parseable outputs (JSON, CSV, delimited lists) from ChatGPT API, then parsing those outputs in Python code for downstream processing. The course demonstrates how to craft prompts that reliably produce structured data and how to handle parsing failures gracefully.

Teaches how to adjust ChatGPT API parameters (temperature, top_p) to control the variability and creativity of outputs. The course explains the trade-off between deterministic, consistent responses (low temperature) and diverse, creative responses (high temperature), with guidance on selecting appropriate values for different use cases.

Teaches best practices for handling failures in ChatGPT API calls, including timeout handling, rate limit management, and fallback strategies. The course demonstrates how to structure Python code that gracefully handles API errors without crashing, with patterns for retry logic and alternative response strategies.

Teaches strategies for reducing API costs by optimizing prompt design, managing token usage, and selecting appropriate model variants. The course covers how to write concise prompts that achieve results with fewer tokens, when to use cheaper models vs more capable ones, and how to measure and track token consumption.