Unit-1(Foundations of AI & Prompt Engineering)

PART 1 — AI BASICS & AI LITERACY

1. What Is Artificial Intelligence?

Artificial Intelligence (AI) is presented in the source material as the simulation of human intelligence processes by machines. AI can learn from data, solve problems, support decisions, understand language, recognize images and speech, and make predictions.

In practical terms, AI is used to automate tasks, analyze information and enhance productivity.

Key idea: AI is not limited to chatbots. It covers a broad family of systems that perform intelligent tasks.

2. Why AI Matters

Automates repetitive work and frees time for creative and strategic tasks.

Processes large amounts of information quickly.

Finds patterns and insights in complex datasets.

Supports data-driven decision-making.

Personalizes recommendations, content and services.

Supports complex applications such as logistics optimization and scientific discovery.

3. A Short AI Timeline

1950 — Alan Turing and the Turing Test.

1956 — John McCarthy and the Dartmouth Conference; the term Artificial Intelligence was formally introduced.

1980s — Expert systems became commercially popular.

2000s — Big data and increased computing power accelerated machine learning and deep learning.

Present — AI is used across healthcare, education, finance, transportation, entertainment and daily-life tools.

4. Types of AI by Capability

Narrow AI (Weak AI): designed for specific tasks; this is the category represented as existing today.

General AI (Strong AI): theoretical system capable of broad human-like intellectual performance across domains.

Super AI (Superintelligence): conceptual system that would surpass human intelligence across fields.

Do not confuse 'theoretical' or 'conceptual' categories with currently available systems.

5. Types of AI by Functionality

Reactive Machines — respond to current input without using past interactions.

Limited Memory AI — uses stored past data to improve current decisions; the source describes most modern AI applications in this category.

Theory of Mind AI — a research/theoretical concept involving understanding human emotions, beliefs and intentions.

Self-Aware AI — hypothetical AI with consciousness and self-identity.

6. AI, Machine Learning and Deep Learning

The relationship is hierarchical: Artificial Intelligence is the broadest field; Machine Learning is a subset of AI; Deep Learning is a subset of Machine Learning.

7. AI in Everyday Life

Education: personalized learning, automated grading, AI tutors and adaptive curriculum design.

Healthcare: disease prediction, medical imaging, drug-discovery assistance and patient monitoring.

Banking: fraud detection, risk assessment, robo-advisors and credit scoring.

Transportation: navigation, autonomous vehicles, traffic management and fleet optimization.

Entertainment: recommendations, gaming opponents and media generation.

Retail and business: inventory management, customer-service bots, demand forecasting and price optimization.

8. AI Literacy

AI literacy is the ability to understand, evaluate and responsibly use AI technologies in everyday and professional contexts.

Understand AI concepts.

Use AI tools effectively.

Evaluate AI outputs for accuracy, bias and misleading information.

Apply ethical awareness around fairness, privacy and accountability.

Use critical thinking rather than accepting AI responses automatically.

9. AI Limitations & Ethics

AI quality depends heavily on data quality.

AI can produce incorrect outputs, particularly in complex or ambiguous situations.

Biases in training data can be reflected or amplified.

AI does not possess genuine human emotions or human common sense.

AI cannot guarantee 100% accuracy.

Responsible use includes privacy, fairness, transparency, accountability and safety.

Best practice: verify important information, protect sensitive data, respect copyright, follow organizational policies and retain human judgment.

PART 2 — PROMPT ENGINEERING MASTERY

10. What Is a Prompt?

A prompt is an instruction, question, command or information provided to an AI system to generate a response. The source emphasizes that vague prompts often force AI to make assumptions.

Example: “Explain Artificial Intelligence.” is valid, but it does not specify audience, length or format.

11. The Five Core Prompt Components

Role — who the AI should act as.

Task — the action the AI should perform.

Context — background information needed to complete the task.

Format — how the response should be presented.

Constraints — limits such as word count, scope, exclusions or reading level.

12. Role-Based Prompting

Role-based prompting asks AI to adopt a particular professional perspective or persona. The source notes that a role can influence vocabulary, tone, perspective, explanation level, recommendations and style.

Example: “Act as an experienced math teacher and explain exponents to Grade 8 students.”

13. Task & Context

A strong task begins with a clear action such as write, draft, summarize, analyze, compare, explain, rewrite, translate, organize, extract or classify.

Context supplies the background needed to perform the task correctly, such as audience, dates, deadlines, locations, purpose, existing information, constraints, policies or previous communication.

14. Prompting Strategies

Zero-Shot — ask AI to perform a task without examples. Quick and useful for common tasks.

One-Shot — provide one example to establish the expected pattern.

Few-Shot — provide multiple examples, typically 3–5 in the source material, to strongly guide pattern recognition.

Role-Based — specify a professional or contextual role.

Step-by-Step / Chain-of-Thought prompting — ask for a structured sequence for problems where intermediate steps are useful.

15. Structured Outputs

Prompting controls not only what AI says, but also how it presents information.

Tables — useful for comparisons, schedules, deadlines and responsibilities.

Bullet points — useful for summaries, instructions and action items.

Checklists — useful when the user must track completion.

Headings — useful for longer content and navigation.

A single prompt can request several formats in sequence—for example, a table, followed by bullet-point summary and then a checklist.

16. Audience, Tone and Length

Students — simple and friendly language.

Faculty — professional and respectful language.

Senior management — concise, formal and executive language.

Public — avoid jargon and explain terms clearly.

Tone can be formal, professional, polite, friendly, concise, neutral, persuasive, respectful, academic or official.

Length can be controlled with explicit limits such as 100 words, five bullet points, one paragraph or approximately 500 words.

17. Include, Exclude and Restrict

A strong prompt explicitly states what must be included and what should be avoided. Examples include including dates, venues, deadlines and contact details while excluding complicated vocabulary, unnecessary introductions, repetition or emojis.

18. Iterative Prompting

Prompting does not have to be perfect on the first attempt. The source presents prompting as an iterative conversation: generate, review, refine and generate again.

Think of prompting as a cycle rather than a single command.

19. Common Prompting Mistakes

Being too vague.

Providing no context.

Failing to specify the audience.

Failing to specify the output format.

Putting too many unrelated tasks into one prompt.

Using ambiguous language such as “Make it better.”

20. Reusable Prompts & Prompt Libraries

Successful prompts can be saved as reusable templates. Keep fixed information—such as tone, greeting, structure and closing—in the template, and replace changing information such as recipient, date, time, purpose, deadline and location.

Meeting invitations

Leave approvals

Examination notices

Fee reminders

Applicant-status emails

Internal memos

Student support replies

Policy communication

Follow-up emails

Official announcements

21. Responsible Prompting

Do not unnecessarily enter passwords or login credentials.

Do not expose private student records or sensitive employee information.

Do not share confidential institutional information without authorization.

Verify important information before acting on it.

Respect copyright and intellectual property.

Follow organizational policies and regulations.

Maintain human oversight.

Prompt quality improves relevance; it does not guarantee factual accuracy.

PART 3 — INTRODUCTION TO AI AGENTS

22. What Is an AI Agent?

The source defines an AI agent as an LLM-powered system capable of autonomously pursuing goals through reasoning, memory, tool use and action loops.

Goal-oriented — works toward a specific objective.

Autonomous — operates with minimal human supervision.

Multi-step planning — breaks larger tasks into smaller steps.

Tool usage — can use APIs, databases and search engines.

Memory — can retain current and past interaction information.

Continuous looping — plans, acts, observes and reflects.

23. LLM vs. AI Agent

LLM: responds to prompts, generates text, reacts to input and may operate statelessly in a single response.

AI Agent: has goals, plans multi-step tasks, uses tools and APIs, maintains memory, acts through workflows and loops continuously.

Conceptual distinction: a chatbot can answer “What is the weather?”; an agent can use a weather API, take follow-up actions and continue a workflow.

24. The Five-Layer Agent Architecture

Perception — receives text, images, audio, APIs and files.

Reasoning — analyzes information, plans steps and decides actions.

Meta-Reasoning — evaluates thinking and self-corrects plans.

Memory — stores and retrieves working, episodic and semantic memory.

Action & Evolution — executes tasks, uses tools and learns from feedback.

25. Memory Systems

Working Memory — short-term information active during the current task/session.

Episodic Memory — records past interactions and events for historical continuity.

Semantic Memory — stable factual knowledge such as policies, manuals and FAQs; often associated with RAG systems.

Memory is presented as the backbone of agent continuity.

26. The ReAct Loop

Agents operate through a continuous cycle rather than a single response. The source describes the cycle as Goal → Plan → Act → Observe → Reflect → Re-plan → Repeat until complete.

Adaptive behavior

Error correction

Autonomous workflows

27. Tool Use

Tools transform an AI system from a text generator into a system capable of taking actions.

Web search and browser/search APIs

Code execution

Database access

External APIs

File handling

Other specialized agents

28. Agent Design Canvas

Purpose — What is the goal?

User Persona — Who will use the agent?

Inputs — What information will it receive?

Outputs — What results will it produce?

Tools Available — Which external systems can it access?

Memory Strategy — How will information be stored?

Constraints — Which actions are forbidden?

Success Metric — How will success be measured?

Before building an agent, define its purpose, boundaries, information flow and success criteria.

29. Multi-Agent Systems

When one agent is not enough, multiple specialized agents can collaborate on a complex workflow.

Supervisor Agent — coordinates the workflow.

Research Agent — collects information.

Analyst Agent — analyzes data.

Writer Agent — creates reports.

Critic Agent — reviews outputs.

The source highlights specialization, parallel execution, higher accuracy and scalability as benefits.

30. Current Agent Limitations

Long-horizon planning — extended workflows can cause loss of focus or context.

Tool failure recovery — unexpected API/system errors can derail workflows.

Doom loops — agents can repeat incorrect actions without detecting failure.

Uncertainty awareness — agents may proceed confidently despite missing or incorrect information.

Human oversight remains essential for monitoring, ethical control, error correction and final decision-making.

PART 4 — INTEGRATED PRACTICE

31. Practice Activity: Build a Strong Prompt

Scenario: You need an AI-generated announcement for a university workshop.

Weak prompt: “Write an announcement about the workshop.”

Improve it by specifying:

Role

Task

Context

Audience

Format

Tone

Length

Required information

Information AI must not invent

32. Practice Activity: Prompt Refinement

Start with: “Explain AI.”

Add a target audience.

Set a word limit.

Request real-world examples.

Specify bullet-point format.

Ask for simple language.

Review the output and issue a second refinement instruction.

33. Practice Activity: Design an AI Agent

Design a hypothetical “Student Support Agent” using the Agent Design Canvas.

Purpose: define the problem the agent should solve.

User Persona: identify the intended users.

Inputs: list information the agent receives.

Outputs: define what it produces.

Tools: identify permitted external systems.

Memory: decide what should persist.

Constraints: specify prohibited actions.

Success Metric: define measurable success.

34. Reflection Questions

Why does a vague prompt force AI to make assumptions?

When would few-shot prompting be more useful than zero-shot prompting?

Why should a prompt specify both audience and output format?

Why does good prompting not guarantee factual accuracy?

What makes an AI agent different from a single LLM response?

How do working, episodic and semantic memory differ?

Why is the Observe → Reflect → Re-plan stage important in an agent loop?

What risks arise when an agent has access to external tools?

Why is human oversight still required even when agents are autonomous?

35. Final Knowledge Check

Which prompt is strongest?

A. Explain AI.

B. Explain AI simply.

C. Explain AI to college students in 150 words with three workplace examples and bullet points.

D. Explain AI in detail.

Which memory type stores stable factual knowledge such as manuals and policies?

A. Working Memory

B. Episodic Memory

C. Semantic Memory

D. Reactive Memory

Which sequence best represents the ReAct loop?

A. Prompt → Answer → Stop

B. Goal → Plan → Act → Observe → Reflect → Re-plan

C. Input → Output → Delete

D. Search → Write → Send

What is the main purpose of constraints in a prompt?

A. Make every prompt longer

B. Limit or control the requested output

C. Remove context

D. Replace the task

Which statement is correct?

A. Better prompting guarantees truth

B. AI agents never need human oversight

C. Tools allow agents to interact with external systems

D. Semantic memory is only temporary

Answer Key: 1-C, 2-C, 3-B, 4-B, 5-C

36. Course Summary

AI simulates aspects of human intelligence and is used across many sectors.

AI literacy requires understanding capabilities, limitations, evaluation and ethical use.

Effective prompting is built from Role, Task, Context, Format and Constraints.

Examples, audience, tone, length and output structure improve control over results.

Prompting is iterative: generate, review, refine and verify.

AI agents extend LLM capabilities with goals, planning, memory, tools and action loops.

The agent architecture combines Perception, Reasoning, Meta-Reasoning, Memory and Action & Evolution.

ReAct enables continuous Goal/Plan/Act/Observe/Reflect/Re-plan behavior.

Multi-agent systems divide complex work among specialized agents.

Human oversight remains critical for accuracy, safety, ethics and final decisions.

Source Basis

This course was developed from the two uploaded presentations: “Unit 1: Foundations of AI & Prompt Engineering” and “Module 3 — Introduction to AI Agents.” The course preserves the terminology and instructional framing presented in those materials and does not claim additional source-specific facts beyond them.