Customer support
Find relevant knowledge, answer a question, and hand off when needed.
Why Applied AI?
Guide AI with clear context, use it in software work, and check the result.
Where Applied AI earns its place.
Prompt engineering helps you define a task, supply relevant context and examples, and explain what a useful answer should contain. In software work, that can help with tests, code explanations, documentation, and support workflows. The result still needs an engineer's review, testing, and judgment.
Developer tools
Suggest code in the editor while a developer remains in control.
Document workflows
Extract, summarize, or search information with checks on the result.
Build a support assistant that knows when to stop.
A product team wants an assistant to answer questions from its help articles. Some questions are covered clearly; others involve missing information, account-specific data, or a sensitive issue that belongs with a person.
The constraint
The assistant should use the available material, avoid inventing an answer, and hand off cases it cannot handle safely.
Write the task and boundaries
Create a prompt that states the role, allowed sources, answer format, and when to say it cannot answer.
Test real cases
Build a small evaluation set with clear, ambiguous, missing-source, and sensitive questions.
Add a useful fallback
Show the source used and route weak or sensitive answers to a human-supported path.
What you would produce
A prompt, a small working prototype, an evaluation sheet, and examples of when the assistant should abstain.
Try an engineer’s decisionIf the model gives a confident answer but the help articles contain no supporting information, should the app display it?
No. Confidence in the wording is not evidence. The workflow should withhold unsupported answers, ask for clarification or hand off, and record the case for evaluation.
This is an illustrative task. The scope of a module is agreed before enrolment.
What you would learn by doing it.
Each part of the exercise connects a tool or concept to a decision you would make on a software team.
Prompt engineering
Give the model a precise task, useful context, examples, and boundaries.
You can improve a workflow deliberately rather than hoping a vague prompt works.Evaluation
Check answers against cases with known expectations, including failures.
You can tell whether a change helps or merely sounds better.Application integration
Connect the model to a small app with source handling and fallback behavior.
AI becomes a reviewable product feature under developer control.How real teams use Applied AI.
Published product and engineering sources show the kind of work this skill supports.
Intercom
Fin uses support content to answer customer questions across chat and email and can escalate to people.
Read the source: Intercom product docsGitHub
Copilot provides code completion suggestions inside developers' editors.
Read the source: GitHub engineeringThese organizations are examples of technology in use. They are not affiliated with PixyDelta.
Find your starting point.
Work through a practical problem with a working engineer and review the decisions behind your result.
Safikul IslamFull-stack developer · In-person sessions