ChatGPT Prompt Engineering for Developers

Section 1: How LLMs Work from a Developer's Point of View

• Tokens, context windows and their effect on input and output length
• Why output varies, and why it must always be checked
• Reasoning models and tasks that need several steps of thought
• Key limits: hallucination, knowledge cutoff and APIs that do not exist

Section 2: Principles of a Good Prompt

• The parts of a prompt: role, context, task, constraints and output format
• Using delimiters to separate instructions from data
• Zero-shot, few-shot and choosing good examples
• Breaking complex work into steps and having the model check its own output
• Lab: improve a learner's existing prompt and compare the results before and after

Section 3: Lab: Prompts for Writing and Improving Code

• Generate code from requirements, stating the language, version and team standards
• Refactoring and explaining legacy code
• Translating code across languages and versions with care
• Lab: have ChatGPT build and refine DevAssist code against requirements

Section 4: Lab: Prompts for Review, Testing and Debugging

• Code review with a checklist covering security, performance and readability
• Generate unit tests and edge cases from code or requirements, then check the coverage is real
• Analyse error logs and stack traces by asking step-by-step questions
• Lab: review code and trace the cause of bugs from sample bug reports

Section 5: Prompts for Documentation and Using AI in the Organisation

• Generate READMEs, API documentation and commit messages
• Write SQL and regular expressions with explanations
• Check AI-generated code before use: tests, dependencies and licences
• Data that must never be sent to ChatGPT, and team guidelines

Section 6: Workshop: The DevAssist Prompt Library

• Build at least 5 prompts: code generation, code review, unit tests, bug analysis and documentation
• Test them against sample code and bug reports
• Record a before-and-after comparison as the prompts are improved
• Store them as templates the team can reuse

Section 7: Lab: Getting Started with the OpenAI Responses API

• Request structure: instructions, input and model
• Choosing a GPT-6 family model for quality, speed and cost
• Handling API keys, timeouts, retries and rate limits
• Multi-turn conversations and streaming output
• The Assistants API has been shut down, and how to move to the Responses API

Section 8: Structured Outputs

• Define a JSON schema so output matches what the system needs
• Use Pydantic to define the schema and validate results
• Handle refusals and incomplete data
• Lab: code review results as JSON (issue, severity, recommendation)

Section 9: Function Calling (Tools)

• Define tools and parameters for the model to choose from
• The loop: the model requests a tool, the system runs the function and returns the result
• Design tools safely: limit permissions and validate every argument
• Lab: a tool that reads related files in the sample project

Section 10: Evaluation: Measuring Prompt Quality Systematically

• Build a set of test cases with clear pass criteria
• Rule-based checks and LLM-as-a-judge scoring
• Compare prompts and models with pytest while tracking cost
• Lab: measure two versions of the code review prompt

Section 11: Security and Working with Other Models

• Direct and indirect prompt injection, and how to defend against it
• Keeping secrets and personal data out of prompts in line with PDPA
• Principles that apply to other LLM providers, and what to adjust when switching models
• Lab: attack DevAssist with prompt injection, then put defences in place

Section 12: Master Workshop: DevAssist CLI

• A Python CLI that takes a code file and reviews it through the Responses API
• Return results as JSON to a schema, and use function calling to read related files
• An evaluation suite in pytest that also passes the prompt injection tests
• Present a quality comparison of each prompt version