Dr. Thamizharasi Ayyavoo · Ph.D · Teaching 16+ yrs · IT 6+ yrs · AI Enthusiast
PromptCraft Pro
AI Prompt Engineering Portal
Select your role to generate a complete, ready-to-run AI productivity session — real prompts, live demos, and workflows by Dr. Thamizharasi.
Practical Focus
Real-World Use Cases
No theory. Every use case mapped to actual daily tasks in your IT role.
Copy-Paste Ready
10 Prompt Templates
Prompts employees can use immediately — structured, tested, role-specific.
Enterprise Safe
Risk & Safety Section
Covers hallucinations, confidentiality, security — everything teams need to know.
SECTION 04
Role-Specific Daily AI Use Cases
SECTION 05
Top 10 Practical AI Prompts
SECTION 06
Live Demo Using First Prompt
LIVE
Use the First Prompt from the Prompts Tab
This demo uses the first prompt from the selected role's Prompts tab. Select a role on the Home page, then use Prompt 1 below as the live demonstration.
🎯 First Prompt for Selected Role
PROMPT 1 DEMO
Generate a session from the Home tab to load Prompt 1 here.
🎬 How to Present This Live
Step 1
Select Role
Choose your team role on the Home page
Step 2
View Prompt 1
It appears above, loaded automatically
Step 3
Paste in ChatGPT / Claude
Copy and generate a live answer
Step 4
Explain Output
Show how prompt quality affects the result
SECTION 07
Live AI Prompt Demo
LIVE
Live AI Demo — Select a Role First
Select your role on the Home page, then use the buttons below to paste the first prompt directly into ChatGPT or Claude for a live demonstration.
🎯 Live Demo Prompt
PROMPT 1 — LIVE SESSION
Generate a session from the Home tab to load the first prompt here.
Step 1
Select Your Role
Choose your team role on the Home page and click Generate Session.
Step 2
Load the Prompt
The first prompt for your role appears here automatically, ready to copy.
Step 3
Run Live
Paste into ChatGPT or Claude during the session and discuss the AI output.
SECTION 01
Basics of Prompt Engineering
FOUNDATION
Understanding Prompts & Prompt Engineering
Prompt engineering is the process of designing clear and structured instructions for AI systems to generate accurate and useful results.
Prompt
Instruction to AI
A prompt tells the AI what task to perform and how to respond.
Goal
Clear Requirement
Better prompts lead to better accuracy, consistency, and automation.
Business Impact
Enterprise Productivity
Prompt engineering improves developer productivity and reduces AI cost.
📌 Components of a Good Prompt
| Component | Description |
|---|---|
| Role | Tell AI who it should act as |
| Task | Clearly explain the work to be done |
| Context | Provide project details and requirements |
| Format | Specify output style such as JSON or table |
| Constraints | Add rules like no assumptions or max word count |
🔬 Prompt Anatomy — Technical Breakdown
ANATOMY OF A PRODUCTION-GRADE PROMPT
[ROLE] Act as a senior FastAPI developer and security architect.
[CONTEXT] I am building a customer login API for a retail CRM serving 14 stores.
Stack: Python 3.11, FastAPI, PostgreSQL, Redis, JWT (RS256).
[TASK] Create a secure login endpoint with: JWT RS256 authentication,
bcrypt password verification, Redis session management,
account lockout after 5 failed attempts, and structured error responses.
[FORMAT] Output: (1) FastAPI route code, (2) Pydantic request/response models,
(3) Redis session key schema, (4) error response structure.
[CONSTRAINTS] No hardcoded secrets. No synchronous DB calls in async endpoints.
Parameterised queries only. Include docstrings. Python type hints throughout.
⚙️ How LLMs Process Your Prompt
| Stage | What Happens | Why It Matters for You |
|---|---|---|
| Tokenisation | Your text is split into tokens (~4 characters each). "FastAPI" = 2 tokens. | Longer prompts cost more. Every word is measured. |
| Attention | The model weights relationships between every token in your prompt simultaneously. | Context at the start and end of a prompt gets highest attention. Bury critical instructions in the middle and the model may miss them. |
| Context Window | The model can only see tokens within its context limit (e.g. 200K for Claude 3.5, 1M for Gemini 1.5 Pro). | Exceeding the window silently truncates your input. Large codebases, full PDFs, and long chat history all compete for this space. |
| Temperature | Controls randomness in token selection. Low (0.0–0.2) = deterministic. High (0.8–1.0) = creative. | Use low temperature for code, SQL, structured output. Use higher temperature for brainstorming and creative writing. |
| System Prompt vs User Prompt | System prompts set persistent behaviour rules. User prompts contain the specific task. | For enterprise AI apps, your role, rules, and constraints belong in the system prompt — not repeated per request. |
| Stop Sequences | Tokens that tell the model to stop generating (e.g. "###", "---END---"). | Use stop sequences in API calls to control output length and prevent runaway generation in structured output flows. |
🧠 Advanced Prompting Techniques
Chain of Thought
Step-by-Step Reasoning
Add "Think step by step before answering" for complex logic, debugging, or multi-step analysis. Forces the model to show its reasoning and significantly improves accuracy.
Few-Shot Prompting
Lead by Example
Provide 2–3 input/output examples inside your prompt before the actual task. The model pattern-matches your examples — ideal for structured output, custom formats, and code style.
ReAct Pattern
Reason + Act + Observe
Used in AI agents: model reasons about the task, selects a tool/action, observes the result, then reasons again. Powers LangGraph, AutoGPT, and production AI agents.
Prompt Chaining
Break into Sub-Tasks
Instead of one giant prompt, chain outputs: Prompt 1 → analyse requirement → Prompt 2 → generate user stories → Prompt 3 → write test cases. Better quality at each step.
Negative Constraints
Tell It What NOT to Do
Explicit exclusions are as powerful as instructions: "No f-strings with user input. No hardcoded secrets. No synchronous calls in async context." LLMs respect clear guardrails.
Structured Output
Force Parseable Responses
Add "Respond ONLY in valid JSON. No preamble. No markdown." to make AI output machine-readable. Essential for any AI pipeline, automation, or API integration.
✅ Production-Ready Example Prompt
GOOD — ENTERPRISE GRADE
Act as a senior FastAPI developer. Create a secure customer login API using JWT RS256 authentication, PostgreSQL with parameterised queries, bcrypt password verification, Redis session storage, account lockout after 5 failed attempts, and Pydantic request/response models. Output: (1) FastAPI route code with async pattern, (2) Pydantic models, (3) Redis key schema, (4) error response structure. Constraints: no hardcoded secrets, no synchronous DB calls in async endpoints, full Python type hints throughout.
SECTION 10
LLM Token Optimization & Cost Control
NEW
Input Tokens, Output Tokens, Context Window & Enterprise AI Costing
This tab explains how LLM cost is calculated and how teams can reduce cost while improving speed, quality, and scalability.
Input Tokens
What You Send to the Model
Includes system prompt, user question, chat history, RAG chunks, uploaded text, schemas, and examples.
Output Tokens
What the Model Generates
Includes answers, summaries, code, JSON, reports, explanations, and generated documents.
Cost
Token Usage = API Cost
More input and output tokens increase cost, latency, and infrastructure load.
📌 Token Cost Formula
BASIC COST MODEL
Total Cost = Input Token Cost + Output Token Cost
Input Token Cost = input_tokens × model_input_price
Output Token Cost = output_tokens × model_output_price
Important: Output tokens are often more expensive than input tokens.
💼 Costing Factors to Compare Across LLMs
| Factor | Why It Matters |
|---|---|
| Input token price | Cost of prompt, system instructions, RAG chunks, chat history. |
| Output token price | Cost of generated answer; often more expensive than input. |
| Context window | How much text, document content, code, schema, or chat history can be sent at once. |
| Max output tokens | Maximum answer size the model can generate in one response. |
| Prompt caching | Reduces cost for repeated system prompts or repeated context. |
| Batch processing | Cheaper for non-real-time jobs such as nightly summaries, report generation, and document processing. |
| Latency | Important for chatbots, live applications, customer support, and real-time copilots. |
| Tool calling / function calling | Needed for agents, API workflows, database tools, automation, and multi-step task execution. |
| Multimodal support | Text, image, audio, PDF, video, screenshots, and document understanding capabilities. |
| Data privacy / enterprise controls | Important for company data, client data, production code, compliance, and access control. |
| Model routing | Use cheaper models for simple tasks and powerful models for complex reasoning, architecture, or code analysis. |
🧠 What Consumes Input Tokens?
| Component | Example | Optimization |
|---|---|---|
| System Prompt | Long role instructions and rules | Compress prompt, remove repeated instructions |
| Chat History | Previous 50 conversation turns | Use summarized memory instead of full history |
| RAG Context | 10 retrieved chunks from documents | Use top 3–5 relevant chunks only |
| Database Schema | Full schema with 100 tables | Send only relevant tables and columns |
| Examples | Many few-shot examples | Keep only high-value examples |
⚠️ Common Token Waste Problems
❌ COSTLY
Sending full PDFs, full chat history, full database schema, and long system prompts for every request.
Impact: high cost, slow response, model confusion, poor scalability.
✅ OPTIMIZED
Send compressed prompt + summarized memory + top relevant RAG chunks + constrained output format.
Impact: lower cost, faster response, better focus, easier enterprise scaling.
⚡ Optimization Techniques
Prompt Compression
Shorter System Prompts
Remove repeated rules, long explanations, and unnecessary examples from every request.
RAG Filtering
Retrieve Only Relevant Chunks
Use metadata filters, top-k tuning, and reranking to avoid sending irrelevant context.
Structured Output
Control Response Size
Ask for JSON, tables, or 5 bullets instead of open-ended long explanations.
Memory Summary
Summarize Conversation State
Replace long chat history with compact user intent, decisions, and pending context.
Caching
Avoid Repeated LLM Calls
Cache static answers, embeddings, summaries, and repeated prompt results using Redis.
Model Routing
Use Right Model for Right Task
Small model for classification, larger model for reasoning, architecture, or complex code.
📊 Enterprise Monitoring KPIs
| KPI | Why It Matters | Action |
|---|---|---|
| Avg Input Tokens | Detect prompt/context bloat | Compress prompt and reduce retrieved chunks |
| Avg Output Tokens | Control response verbosity | Add output length limits and structured format |
| Cost per Request | Understand unit economics | Set budget alerts and quotas |
| Cost per User / Team | Identify expensive workflows | Apply model routing and caching |
| Token Spikes | Detect abuse or accidental full-document sends | Add max token caps and input validation |
| Cache Hit Ratio | Measure saved LLM calls | Improve semantic cache and TTL strategy |
🎬 Live Demo Flow
Before
Huge Prompt
Full history + full schema + full docs
AI Analysis
Token Audit
Find wasteful input/output tokens
After
Optimized Prompt
Compressed prompt + top-k context
Result
Lower Cost
Faster response and cheaper scaling
LIVE DEMO PROMPT
Act as an LLM cost optimization expert.
Analyze this AI application prompt flow and identify token waste:
- System prompt: 2500 tokens
- Chat history: 6000 tokens
- RAG context: 8000 tokens
- User query: 100 tokens
- Output response: 1500 tokens
Provide:
1. Total tokens per request
2. Main cost drivers
3. Optimization plan
4. Before vs after token estimate
5. Enterprise monitoring KPIs
✅ Closing Message
Prompt Engineering is also Cost Engineering
In enterprise AI systems, better prompts do not only improve answer quality.
They reduce token usage, control API cost, improve latency, and make AI applications scalable.
SECTION 02
Reusable Prompt Design Patterns
PATTERNS
Prompt Structures Employees Can Use Daily
Prompt patterns are reusable structures that help employees get consistent, high-quality AI output.
Role Prompting
Assign Expert Role
Ask AI to act as a senior developer, QA lead, architect, HR recruiter, or delivery manager.
Context Prompting
Give Project Background
Mention tech stack, business domain, constraints, environment, and expected users.
Format Prompting
Control the Output
Request output as table, JSON, checklist, Jira story, RCA report, or code block.
Few-Shot Prompting
Show Examples
Provide 1–3 examples of input and expected output so AI follows your pattern.
Constraint Prompting
Set Boundaries
Specify limits: max words, no assumptions, no confidential data, cite gaps clearly.
Review Prompting
Ask AI to Validate
Ask AI to check its own answer for missing risks, edge cases, and contradictions.
📌 Master Prompt Pattern
ROLE + TASK + CONTEXT + FORMAT + CONSTRAINTS
Act as a [ROLE].
Task:
[Clearly describe what you want]
Context:
[Project, stack, users, business domain, environment]
Output Format:
[Table / JSON / Jira story / code / checklist]
Constraints:
[Security limits, word limit, assumptions, validation rules]
⚖️ Bad Prompt vs Pattern-Based Prompt
❌ BAD
Fix this code and make it better.
No role, no goal, no stack, no output format.
✅ GOOD
Act as a senior Node.js backend engineer. Review this Express API for security, performance, validation, and error handling. Return findings in a table with severity, issue, fix, and corrected code.
Role + task + quality criteria + output format.
SECTION 09
AI Hallucination Prevention
CRITICAL
How to Reduce Wrong, Fake, or Overconfident AI Answers
Hallucination happens when AI gives confident answers that are incorrect, invented, outdated, or unsupported.
Grounding
Use Source Context
Give AI verified documents, schemas, logs, or retrieved RAG chunks instead of asking from memory.
Uncertainty
Force AI to Say “Not Sure”
Ask AI to clearly state assumptions, missing information, and confidence level.
Verification
Ask for Validation Steps
For code, SQL, cloud, and security: ask how to verify the answer before production use.
Schema Control
Provide Exact Schema
For SQL/API work, provide exact table columns or API contract so AI does not invent fields.
No Guessing
Block Unsupported Claims
Tell AI: “Do not invent facts. If data is missing, list what is missing.”
Cross Check
Use Review Prompt
Ask AI to review its own answer and identify possible inaccuracies or weak assumptions.
🛡️ Hallucination-Safe Prompt
ENTERPRISE SAFE PROMPT
Act as a careful technical reviewer.
Use only the information provided below.
Do not invent API names, table columns, library functions, prices, dates, or policies.
If something is missing, write “Information not provided.”
For every recommendation, include:
- Evidence from the given context
- Assumption, if any
- Verification step before production use
Context:
[paste logs / schema / requirement / document here]
🎬 Live Demo Idea
Step 1
Ask Vague Question
AI may invent details
Step 2
Add Source Context
Provide schema/logs/docs
Step 3
Add No-Guess Rule
Force uncertainty handling
Result
More Reliable Answer
Safer for enterprise use
SECTION 08
AI Output Formats Across IT Roles
REFERENCE
Every Output Format Used in IT — When to Use Each and How to Ask for It
The format you request determines whether AI output is immediately usable or needs manual cleanup. Specify the format explicitly in every prompt — AI will match it precisely.
📋 Complete Output Format Reference by IT Role
| Format | Used By | Best For | Prompt Instruction to Use |
|---|---|---|---|
| JSON | AI Developer, Backend Dev, DevOps | API responses, agent tool outputs, config files, structured data pipelines, automation payloads | "Return ONLY valid JSON. No preamble. No markdown fences." |
| Markdown Table | BA, QA, PM, Delivery Manager, TA | Test cases, RAID logs, story backlogs, comparison matrices, sprint plans, risk registers | "Format as a markdown table with columns: [col1 | col2 | col3]" |
| Numbered List | All roles | Step-by-step runbooks, ordered requirements, installation guides, release checklists, onboarding steps | "Return as a numbered list. Each item one line. No sub-bullets." |
| Code Block | AI Dev, Frontend Dev, Backend Dev, DevOps, Network Eng | Working code, SQL queries, Dockerfile, YAML configs, shell scripts, IaC templates | "Return only the code inside a single code block. Language: Python. No explanation outside the block." |
| YAML | DevOps, Technical Architect, AI Developer | Docker Compose, Kubernetes manifests, GitHub Actions pipelines, OpenAPI specs, Ansible playbooks | "Output as valid YAML only. Use 2-space indentation. No comments unless requested." |
| OpenAPI / Swagger YAML | Technical Architect, Backend Dev, BA | REST API contracts, endpoint documentation, request/response schemas, integration specs | "Output a complete OpenAPI 3.1 YAML spec with paths, schemas, and example values." |
| Gherkin (Given/When/Then) | QA, BA, PM | Acceptance criteria, BDD test scenarios, UAT scripts, feature file generation | "Write acceptance criteria in Gherkin format: Given [context], When [action], Then [outcome]." |
| Mermaid Diagram | Technical Architect, BA, AI Developer | Flowcharts, sequence diagrams, entity-relationship diagrams, system architecture, state machines | "Generate a Mermaid flowchart / sequenceDiagram / erDiagram. Output only the Mermaid code block." |
| PlantUML | Technical Architect, BA | UML class diagrams, activity diagrams, component diagrams, use case diagrams for formal documentation | "Generate PlantUML code for a [diagram type] diagram. Output only the @startuml...@enduml block." |
| SQL | AI Developer, Backend Dev, BA, DBA | Query generation, schema creation, index recommendations, migration scripts, analytics queries | "Write a SQL query for PostgreSQL. Use parameterised placeholders. No dynamic SQL. Add inline comments." |
| Prose / Executive Summary | Delivery Manager, PM, CTO, HR | RAG status reports, board-level summaries, incident narratives, client emails, PIRs | "Write in professional prose. Max [N] words. Structure: [intro / body / action items]. No bullet points." |
| Bullet Summary | PM, Delivery Manager, HR, UI/UX | Meeting summaries, sprint retrospectives, risk highlights, interview notes, research findings | "Summarise as bullet points. Max [N] bullets. Each bullet one sentence. Group under headers: [Section 1, Section 2]." |
| JIRA-Ready Format | BA, PM, QA, Delivery Manager | Epics, user stories, subtasks, acceptance criteria, story point estimates, sprint assignment | "Format as JIRA-ready output: Epic | Story | Subtask | Acceptance Criteria | Story Points | Priority | Sprint." |
| CSV | PM, HR, BA, DevOps | Bulk data import, test case exports, candidate tracking, SLA logs, sprint metrics for Excel/Sheets | "Output as CSV. First row is headers. No extra text. Wrap fields with commas in double quotes." |
| Terraform / HCL | DevOps, Technical Architect | AWS/GCP/Azure IaC provisioning, reusable modules, state management, environment configs | "Write Terraform HCL for [resource]. Use variables for all hardcoded values. Add description to every variable." |
| Dockerfile / docker-compose.yml | DevOps, AI Developer, Backend Dev | Container build definitions, multi-service local dev, production service orchestration | "Write a production-ready Dockerfile using multi-stage build. Then write the docker-compose.yml with health checks." |
| Regex Pattern | Backend Dev, QA, DevOps, Network Eng | Input validation, log parsing, search-and-replace, firewall rules, URL routing patterns | "Return only the regex pattern. Include a Python snippet showing how to apply it. Explain each group in one line." |
| Pydantic / TypeScript Schema | AI Developer, Backend Dev, Frontend Dev | Request/response models, type-safe data contracts, API validation, frontend interfaces | "Generate a Pydantic v2 BaseModel / TypeScript interface for [data structure]. Include field descriptions and validators." |
| Scorecard / Rubric | HR, Delivery Manager, PM | Interview scorecards, CV screening rubrics, vendor evaluation matrices, code review checklists | "Format as a scoring rubric: Criterion | Weight | 1 (Does Not Meet) | 2 (Partial) | 3 (Meets) | 4 (Exceeds)." |
| ADR (Architecture Decision Record) | Technical Architect, CTO, AI Developer | Documenting technology choices, trade-off analysis, team alignment, audit trail for future decisions | "Write an ADR. Format: Title | Status | Context | Decision | Options (with pros/cons) | Consequences | Risks." |
| Changelog / Release Notes | DevOps, PM, Backend Dev, Frontend Dev | Sprint release notes, API version changelogs, hotfix summaries, customer-facing feature announcements | "Write release notes in Keep a Changelog format: Added | Changed | Fixed | Deprecated | Security. Audience: [technical/non-technical]." |
| Wireframe Description / UI Spec | UI/UX Designer, Frontend Dev, BA | Component state definitions, layout specs, interaction documentation, handoff to developers | "Write a UI spec for [component]. Cover: layout, all states (default/hover/loading/error/empty), interactions, ARIA attributes, responsive breakpoints." |
| RFC Document | Technical Architect, CTO, AI Developer | Proposing architecture changes, API redesigns, process improvements for team review and consensus | "Write an RFC. Format: Summary | Motivation | Proposed Design | Alternatives | Drawbacks | Unresolved Questions | Implementation Plan." |
| Post-Incident Report (PIR) | DevOps, Delivery Manager, PM | Production incident documentation, root cause analysis, corrective actions, stakeholder communication | "Write a PIR. Format: Executive Summary | Timeline (UTC) | 5-Whys Root Cause | Actions Taken | Corrective Actions with Owner and Deadline | Prevention." |
| STRIDE Threat Model | Technical Architect, AI Developer, DevOps | Security architecture reviews, threat identification, risk-ranked mitigations, pen-test prep | "Generate a STRIDE threat model. For each category: specific threat, Likelihood × Impact, mitigation, component owner. End with top 5 by risk score." |
⚡ Format Selection Quick Guide
Machine Consumes It
Use: JSON, YAML, CSV, SQL, Code Block, Regex, Schema
When the output feeds into an API, pipeline, database, script, or automation — the machine needs clean parseable text. Zero prose outside the format.
Human Reviews It
Use: Markdown Table, Numbered List, Bullet Summary, Gherkin, Scorecard
When a BA, QA, PM, or developer reads and validates the output — use formats they recognise from their daily tools: Confluence, Notion, JIRA, Excel.
Stakeholder Presents It
Use: Prose, Executive Summary, ADR, RFC, PIR, Release Notes
When the output goes to a client, board, or cross-team audience — formal prose with clear sections, specific dates, named owners, and no jargon.
🎯 Format Instruction Examples — Copy Ready
JSON — API OUTPUT
Return ONLY valid JSON matching this exact schema. No preamble. No explanation. No markdown fences.
{
"status": "success | error",
"data": {},
"errors": [],
"metadata": { "source": "", "confidence": 0.0 }
}
MARKDOWN TABLE — QA / BA
Return output as a markdown table only.
Columns: ID | Category | Test Scenario | Precondition | Steps | Expected Result | Priority
Start from ID TC-001. Priority values: P1 / P2 / P3. No rows outside this table.
JIRA FORMAT — BA / PM
Format output as JIRA-ready tasks. Structure:
EPIC: [title] — [business objective]
STORY [n]: As a [role], I want [action], so that [benefit].
Acceptance Criteria (Gherkin):
Given [context]
When [action]
Then [outcome]
Story Points: [n] | Priority: High/Medium/Low | Sprint: Sprint [n]
SUBTASK: [technical task] — Owner: [role] | Est: [n] hours
EXECUTIVE PROSE — DELIVERY MANAGER / CTO
Write in professional executive prose. Structure:
1. Overall Status: [RAG: Red / Amber / Green] — one sentence justification
2. Executive Summary: 3 sentences maximum. No jargon.
3. Key Achievements: bulleted, specific, with dates
4. Risks and Issues: each with owner, mitigation, and resolution date
5. Next Period Plan: what delivers by when
Max 300 words total. Audience: non-technical senior leadership.
SECTION 11
Bad Prompt vs Good Prompt
Example 1 — Developer: Code Fix Request
❌ BAD PROMPT
Hi, I have a bug in my FastAPI endpoint. It's been working fine for a while but now it sometimes returns a 500 error. I'm not sure what's causing it. I've tried restarting the server and checking the logs but couldn't find anything obvious. Can you help me fix it? Let me know if you need more information. I'm using Python and PostgreSQL. The error seems to happen more when there are multiple users. Thanks.
Vague symptom ("sometimes 500"), no code, no stack trace, no repro steps, no versions — AI can only guess generically.
✅ GOOD PROMPT
Act as a senior backend engineer. My FastAPI async endpoint POST /api/chat raises a 500 error under concurrent load (5+ users). Stack: Python 3.11, FastAPI 0.110, asyncpg, PostgreSQL 15. Error: "asyncpg.exceptions.TooManyConnectionsError: sorry, too many clients." The endpoint opens a new DB connection per request with no connection pooling. Review this code for the connection handling bug, explain the root cause, and provide the fix using asyncpg connection pool with lifespan context manager. Also identify any other async anti-patterns in the same code.
Role + exact error message + stack versions + reproduction condition + code pattern + specific fix requested = surgical answer in one response.
Example 2 — BA: Requirements Documentation
❌ BAD PROMPT
We had a meeting with the client today and they mentioned they want some improvements to the system. They talked about adding a dashboard and making things faster. There were also some concerns about reports not being accurate. Can you help me write up the requirements from this? The client is a retail company and the project is about 6 months. I want to make it professional and clear for the developers. Please write it in a proper format.
No meeting notes, no actual requirements, no stakeholders named, no feature scope — AI invents plausible-sounding but fabricated requirements.
✅ GOOD PROMPT
Act as a senior Business Analyst. Convert these raw stakeholder meeting notes into a formal BRD section for a retail CRM dashboard module. Client: UAE jewelry retailer, 14 stores. Stakeholders: IT Director (technical), Store Operations Head (non-technical). Raw notes: "need real-time KPI dashboard per store, current Excel reports are 2 days delayed, managers want footfall, revenue, conversion rate on one screen, export to PDF for board meetings, Arabic language support needed, mobile must work on iPhone." Format: Business Objective, Current State Problem, Proposed Solution (numbered functional requirements), Business Rules, Assumptions, Constraints. Flag any ambiguities as open questions.
Role + raw notes provided + stakeholder context + output format + ambiguity instruction = a BRD section ready for sign-off review.
Example 3 — QA: Test Case Generation
❌ BAD PROMPT
I need test cases for the login feature. We want to make sure it works properly and handles errors. The login has a username and password. Write test cases covering different scenarios. We're using a standard web application. Please make it comprehensive and include edge cases. The format should be something the QA team can use. Let me know if you need anything else about the system.
No auth mechanism, no business rules (lockout policy, session timeout), no tech stack — produces 5 generic happy-path tests that miss real failure modes.
✅ GOOD PROMPT
Act as a senior QA engineer. Write a complete test suite for a corporate login module. Auth: Azure AD SSO + local username/password fallback. Rules: MFA required for admin roles, account locks after 5 failed attempts for 15 minutes, JWT session expires after 30 minutes idle, password complexity: 12 chars min with uppercase, number, special char. Generate test cases in this table format — ID | Category | Test Scenario | Precondition | Steps | Expected Result | Priority — covering: positive flows (SSO success, local login success), negative flows (wrong password, locked account, expired session, MFA failure), edge cases (special chars in password, concurrent sessions, token refresh), and security scenarios (SQL injection in username field, brute force detection).
Auth type + all business rules + exact table format + 4 coverage categories = a test suite a developer implements and a manager signs off on.
SECTION 12
Best AI Tools for Your Role
ChatGPT
Prompt testing, coding help, documentation, analysis, and structured output.
Chat
Claude
Long document analysis, code review, reasoning, and enterprise writing.
Reasoning
GitHub Copilot
Inline code completion, test generation, and developer productivity.
IDE
Cursor IDE
AI-native codebase analysis, refactoring, and multi-file editing.
Codebase
Gemini
Large context analysis, multimodal workflows, and cost-sensitive tasks.
Large Context
Langfuse
LLM observability, tracing, cost monitoring, and quality evaluation.
Observability
SECTION 13
AI Risks & Safety — What Every Employee Must Know
⚠️ Hallucination Risk
AI may confidently generate incorrect code, fake APIs, wrong SQL columns, or unsupported claims. Always verify.
🔒 Confidentiality Risk
Do not paste client data, API keys, production credentials, private code, or sensitive documents into public AI tools.
🧪 Validation Required
Treat AI output as a draft. Test code, validate SQL, review security, and confirm business logic before use.
SECTION 14
Quick Prompt Cheat Sheet
01
Act as a [role]. Analyze [task] and return output as [format].
02
Convert this requirement into Jira stories with acceptance criteria and story points.
03
Review this code for bugs, security issues, performance issues, and best practices.
04
Generate test cases covering positive, negative, edge, and security scenarios.
05
Summarize this meeting into decisions, action items, owners, and deadlines.
BUSINESS MODEL
AI Adoption Business Model
MONETIZATION
Turn Prompt Engineering Training into a Sellable AI Productivity Program
This section converts the AI practical session into a business model for corporate training, consulting, subscriptions, and AI adoption support.
🎯 Why Companies Should Invest in AI
Time Saving
Reduce Repetitive Work
Documentation, reports, test cases, Jira tasks, email drafts, code review, and meeting summaries can be accelerated.
Productivity
Improve Team Output
Employees learn how to use AI safely and practically in daily IT workflows.
Governance
Use AI Safely
Training includes hallucination control, prompt injection awareness, confidentiality, and verification practices.
📊 Current Challenges vs AI Benefits
| Team | Current Challenge | AI Benefit | Expected Gain |
|---|---|---|---|
| Developers | Documentation, debugging, code review | Faster analysis, refactoring, unit test generation | 20% – 40% |
| Testers | Test case creation, regression scenarios | Functional, negative, edge, and security test cases | 30% – 50% |
| Business Analysts | Requirements and Jira stories | Epics, stories, subtasks, acceptance criteria | 30% – 60% |
| HR / Talent Acquisition | JD writing and resume screening | JD, scorecards, interview questions, summaries | 40% – 70% |
| Project Managers | Status reports, MoM, planning | Reports, RAID logs, meeting summaries, sprint plans | 20% – 50% |
📈 ROI Calculator Example
SAMPLE ROI CALCULATION
Company Size: 50 Employees
Average Salary: ₹50,000/month
Monthly Working Hours: 160
Cost Per Hour = ₹50,000 ÷ 160 = ₹312/hour
If AI saves only 30 minutes per employee per day:
50 employees × 0.5 hours = 25 hours/day
Monthly Savings:
25 × 22 working days = 550 hours/month
Value Created:
550 × ₹312 = ₹1,71,600/month
Annual Value:
₹20+ Lakhs/year
🎓 Training Payment Plans
Package 1
AI Awareness Session
Audience: Entire IT company
Duration: 2 Hours
Includes: AI basics, ChatGPT, Claude, prompt basics, risks
₹10,000 – ₹25,000
Duration: 2 Hours
Includes: AI basics, ChatGPT, Claude, prompt basics, risks
₹10,000 – ₹25,000
Package 2
Role-Based Productivity Workshop
Audience: Developers, QA, BA, PM, DevOps, HR
Duration: Half Day
Includes: Role-wise prompts, live demos, token optimization
₹25,000 – ₹50,000
Duration: Half Day
Includes: Role-wise prompts, live demos, token optimization
₹25,000 – ₹50,000
Package 3
AI Productivity Accelerator
Duration: 1 Month
Includes: Assessment, training, prompt library, adoption support, ROI measurement
₹75,000 – ₹2,00,000
Includes: Assessment, training, prompt library, adoption support, ROI measurement
₹75,000 – ₹2,00,000
🗓️ AI Productivity Accelerator Plan
Week 1
Assessment
AI readiness and repetitive task analysis
Week 2
Training
Prompt engineering, AI safety, role-wise demos
Week 3
Prompt Library
Custom prompts for teams and departments
Week 4
Adoption Support
Usage review, improvements, ROI measurement
🔄 Subscription Plans
Individual
AI Prompt Library
500+ prompts, templates, cheat sheets, monthly updates, recordings.
₹299/month
₹299/month
Corporate
Team Prompt Portal
Team prompt library, AI updates, monthly webinar, prompt packs, basic support.
₹999/user/month
₹999/user/month
🏢 AI Center of Excellence — Premium Retainer
PREMIUM OFFER
Suitable for 100+ employee companies. Includes AI governance, usage policies, data handling rules, prompt library, AI champions program, monthly workshops, cost optimization reviews, productivity dashboard, and adoption tracking.
Monthly Retainer: ₹25,000 – ₹1,00,000/month
Monthly Retainer: ₹25,000 – ₹1,00,000/month
💰 Additional Revenue Streams
| Service | Target Audience | Pricing |
|---|---|---|
| Resume Optimization Using AI | Job seekers and professionals | ₹999/person |
| LinkedIn AI Branding Workshop | Professionals and job seekers | ₹5,000 – ₹10,000/session |
| AI Interview Preparation | Freshers and experienced candidates | ₹2,000 – ₹5,000/person |
| College Faculty AI Training | Schools, colleges, faculty groups | ₹10,000 – ₹50,000/session |
| Prompt Engineering Consulting | Startups and IT teams | ₹1,000 – ₹3,000/hour |
🎯 Why Choose This Program
Academic Strength
PhD + Teaching Experience
Strong ability to explain AI clearly to beginners, managers, and technical teams.
Industry Strength
AI / ML Project Experience
Practical knowledge in Python, computer vision, GenAI, RAG, dashboards, and enterprise workflows.
Positioning
AI Productivity Consultant
Position as a business productivity consultant, not only as a trainer.
📞 Free AI Adoption Assessment
Start with a Free AI Readiness Assessment
Offer a 30-minute consultation to identify repetitive tasks, AI use cases, cost-saving opportunities, training needs, and the best adoption roadmap for the company.
SECTION 15
Corporate Closing Message
AI is a Productivity Multiplier
Prompt engineering helps teams ask better questions, get better outputs, reduce cost, and use AI safely in enterprise workflows.