The AI must guess the users, calculations, platform, language, privacy and definition of done.
Free visual course / English + हिन्दी / 7 lessons
Prompting is explaining the job clearly.
A prompt is the instruction and context you give an AI so it can produce a useful result. Good prompting is not a secret vocabulary. It is a visible agreement about the goal, inputs, boundaries, output and checks.
Prompt वह निर्देश और जानकारी है जो आप AI को देते हैं, ताकि वह सही और उपयोगी परिणाम दे सके। अच्छे prompting का अर्थ कठिन अंग्रेज़ी नहीं—स्पष्ट उद्देश्य, सही संदर्भ और जाँचने योग्य परिणाम है।
The five-part prompt
From one idea to a result you can check.
Begin with the idea, provide the useful context, name the action, define the output and decide how correctness will be checked.
A prompt is a brief—not an incantation.
The same AI can return very different results when the audience, evidence or output format changes. A useful prompt reduces consequential guessing; it does not make the idea sound unnecessarily complicated.
Ask only the decisions that affect calculations, journeys, privacy and release architecture. Then create a phased, testable Prompt Pack.
Lesson 02 / A short history
Commands became conversations. Conversations gained tools.
- Commands: people gave software exact instructions.
- Chat: language models made instructions conversational and iterative.
- Tools: models began reading sources, running code and calling approved services.
- Agents: systems can plan and act across bounded steps—but still need tests and stopping rules.
There is no single inventor of “prompting.” The terminology and techniques developed across computing, natural-language processing and generative AI.
Five questions make the work visible.
Idea
What do you want?
आप क्या बनवाना चाहते हैं?
Context
What should the AI know?
AI को कौन-सी पृष्ठभूमि जाननी चाहिए?
Instructions
What must it do and avoid?
क्या करना है और क्या नहीं करना है?
Output
What should the result look like?
अंतिम परिणाम किस रूप में चाहिए?
Checks
How will you decide it is correct?
आप कैसे जाँचेंगे कि परिणाम सही है?
Different jobs need different prompt patterns.
Zero-shot
Ask directly when the task and expected format are simple.
Summarise this report in five bullets.
One-shot
Show one example when tone or structure matters.
Use this approved product description as the pattern for the next one.
Few-shot
Use varied examples when the pattern has important edges.
Use these three labelled replies to classify the fourth.
Role + context
Name useful expertise and facts—not a theatrical personality.
Review the supplied configuration as a security engineer.
Structured output
Specify fields, headings or schema when another system will use the result.
Return Severity, Evidence, Impact and Fix.
Prompt chain
Let one checked output become the input to the next bounded stage.
Research → approve facts → outline → draft → verify.
For reasoning work, ask for assumptions, a concise rationale and verification—not private hidden reasoning. A long answer is not automatically a correct answer.
Lesson 05 / One intention, different media
A good image prompt is not a good code prompt.
- Text: audience, evidence, voice and structure.
- Image: subject, composition, light, style and exclusions.
- Video: shots, motion, continuity, sound and duration.
- Audio: voice, pacing, pronunciation, instruments and delivery.
- Code: repository state, architecture, tests, security, migration and rollback.
Lesson 06 / Loops and agents
“Loop” can mean three different workflows.
- Human feedback loop: generate, review and improve.
- Prompt chain: one approved output becomes the next prompt’s input.
- Agent loop: an AI performs bounded actions and tests until it passes or reaches a limit.
Every safe loop needsGoalSaved stateAcceptance testMaximum attemptsCost boundaryStop ruleRollback
Fluent is not the same as verified.
Use authorised web research and cite sources when facts can change.
Do not place secrets, private code or personal data into public research tools.
Use code or a calculator and test known examples instead of trusting prose arithmetic.
Require qualified human review for legal, medical, financial and safety-critical outputs.
Sources and further reading
Start with orientation. Confirm with primary guidance.
- Wikipedia: Prompt engineering ↗Beginner orientation and terminology; verify technical claims against original sources.
- Google AI: Prompt design strategies ↗Iterative design, examples, structure and prompt chaining.
- Chain-of-Thought Prompting ↗Original 2022 research paper.
- ReAct: Reasoning and Acting ↗Original research connecting reasoning with external actions.
Practice / No registration required
Open Prompt Generator →Bring one honest sentence. Let the interview settle the rest.
Choose Basic for a fast guided sequence or Advanced when research and deeper approval are required.