10x Better AI Output in 3 Steps – The Only Prompt Engineering Guide You’ll Ever Need  [Part-05] | FreeLearning365

Bangla AI Series · Post 5 of 20

05

PROMPT ENGINEERING

10X BETTER OUTPUT

"Bad output = bad prompt. Not bad AI."
Claude-এর output ভালো না হলে সমস্যা tool-এ নয় — সমস্যা তোমার prompt-এ। এই post-এ শিখবে exactly কীভাবে prompt লিখলে Claude-এর সত্যিকারের power বের হয়।

🚀 Techniques দেখো📌 Real Scenarios

9X

Average output improvement with right prompting

7

Core prompt engineering techniques covered

15+

Real Bangladesh scenarios with full examples

30+

Copy-paste ready prompt templates

Role Prompting → 3x better outputChain-of-Thought → Complex problems solvedFew-Shot Examples → Consistent style every timeConstraint Prompting → No more bloated responsesIterative Prompting → Perfect output in 3 stepsPersona Prompting → Expert-level responsesTemplate Chaining → Full workflows automatedRole Prompting → 3x better outputChain-of-Thought → Complex problems solvedFew-Shot Examples → Consistent style every timeConstraint Prompting → No more bloated responsesIterative Prompting → Perfect output in 3 stepsPersona Prompting → Expert-level responsesTemplate Chaining → Full workflows automated

The Problem

কেন Same AI
Different Results?

দুইজন মানুষ একই Claude ব্যবহার করে। একজন পায় brilliant, specific, actionable output। আরেকজন পায় generic, vague, useless response। Difference কোথায়? Prompt-এ।

❌ Prompt Without Thinking

"আমার business-এর জন্য email লিখো।"
Claude জানে না: কোন business? কাকে? কী goal? কোন tone? Output হবে unusably generic। "Dear Valued Customer, We hope this email finds you well..." — এই garbage।

"Code fix করো।"
কোন code? কোন error? কোন language? Claude guess করবে। Wrong guess = wasted prompt।

"Content idea দাও।"
Claude তোমার niche, audience, platform, existing content জানে না। দেবে generic list যা Google-এ ১০ বছর ধরে আছে।

✓ Prompt With Intent

"তুমি একজন expert B2B copywriter। আমার Dhaka-based IT consulting firm 'TechBridge BD'-এর জন্য একটা cold outreach email লিখো। Target: RMG factory owners। Goal: free IT audit consultation book করানো। Tone: professional but approachable। 150 words max।"
Claude সব context পেয়েছে। Output হবে specific, usable, on-brand।

"Python 3.11, FastAPI। এই error: [full stacktrace]। Code: [paste]। Root cause বলো, fix করো, same bug prevent করার test দাও।"
Claude জানে exactly কী করতে হবে।

"আমার Bengali tech YouTube channel-এর জন্য ৫টা video idea দাও। Channel niche: AI tools for Bangladeshi professionals। Last 3 videos: [titles]। Audience: 25-40 বছর, Dhaka/Chittagong, working professionals।"
Context = relevant, non-generic ideas।

Output Quality DifferenceGeneric trash vs. Publication-ready content

Claude is as smart as
the question you ask. No more, no less.

— Fundamental truth of prompt engineering

73%

Users never give context in prompts

9X

Improvement with proper Role Prompting

60%

Less prompt usage with Batching technique

3

Iterations to reach perfect output (average)

The Techniques

৭টা Core Technique
যা সব বদলে দেয়

এই ৭টা technique master করলে তুমি ৯৫% use case handle করতে পারবে। প্রতিটায় real Bangladesh examples আছে।

1

Technique 01

ROLE PROMPTING

Claude-কে একটা specific expert identity দাও। "তুমি একজন experienced [PROFESSION]।" এটা Claude-এর response-এর পুরো framing পাল্টে দেয় — সে সেই role-এর perspective থেকে চিন্তা করে।

❌ WITHOUT ROLE: "Sales email লিখো।" Output: Generic "Dear Customer, We are pleased to offer..."

✓ WITH ROLE: "তুমি একজন B2B sales expert যে SaaS companies-এ 15 বছর কাজ করেছে। আমার [PRODUCT]-এর জন্য cold email লিখো। Target: CFO-level executives। Pain point: manual accounting process। Goal: 15-minute discovery call book করানো। Tone: Peer-to-peer, not salesy। 120 words max।"

💡 Role Prompting Power

Role দেওয়া মানে Claude-কে বলছো সে কোন lens দিয়ে দেখবে। একজন "senior financial analyst" এবং একজন "startup founder" একই question-এর আলাদা answer দেবে। উভয়ই valuable — তোমার need অনুযায়ী role choose করো।

Best roles for Bangladeshi professionals:
• "experienced Bangladeshi HR consultant"
• "senior software architect with startup experience"
• "expert Bengali content strategist"
• "experienced RMG export merchandiser"
• "financial advisor specializing in SME Bangladesh"

Real Scenario — Nadia (Freelance Digital Marketer, Dhaka)

Role Prompting-এ Client Pitch Email — Won a ৳3 Lakh Contract

নাদিয়া একজন freelance digital marketer। একটা mid-size fashion brand-কে pitch করতে হবে। আগে generic email পাঠাতো — কোনো reply আসতো না। Role prompting শেখার পর প্রথম email-এই meeting পেলো।

তুমি একজন expert B2C digital marketing consultant যে বাংলাদেশের fashion industry-তে ১০ বছর কাজ করেছো এবং e-commerce brands-এর জন্য ROI-focused campaigns বানিয়েছো। আমার client: "Rang Dhanu Fashion" — Dhaka-based women's ethnic wear brand, 50,000 Facebook followers, currently doing ৳8-10 lakh/month online sales। তারা ঈদের আগে sales ৳20 lakh-এ নিয়ে যেতে চায়। আমার service: Full digital marketing retainer — Facebook/Instagram ads, content strategy, influencer outreach। My rate: ৳1.5 lakh/month। একটা compelling pitch email লিখো যেটা: 1. তাদের specific situation address করে 2. আমার expertise credibility establish করে 3. Clear ROI case দেখায় 4. একটা low-risk first step offer করে (paid audit) 5. Urgency create করে (Eid timing) Tone: Confident consultant (peer-to-peer), not desperate vendor। English with some Bangla brand references। Subject line include করো। 200 words।

Subject: Rang Dhanu Fashion → ৳20L Eid Sales: Here's the Gap I See

Dear [Name],

I've been watching Rang Dhanu Fashion's digital presence for a while — your aesthetic is genuinely beautiful, and your 50K followers are engaged. But I also see a consistent gap between your content quality and your conversion performance that I think is costing you ৳8-12 lakh every month.

Here's what I'm seeing: your organic reach is strong, but your paid acquisition funnel has three specific leaks — I can see them from the outside. With Eid-ul-Fitr now 8 weeks away, this is the last practical window to fix them before your biggest revenue season.

I've helped three women's ethnic wear brands in Dhaka scale from ৳8-10L to ৳18-25L monthly in 90 days — not by spending more, by spending smarter.

I'd like to offer you a ৳25,000 Digital Audit this week — a concrete 7-point analysis of exactly where your ৳2Cr+ annual revenue potential is leaking, with specific Eid campaign recommendations. If you don't find it worth 10x the fee, I'll refund it entirely.

Would Thursday work for a 20-minute call?

Nadia [Your Name]
Digital Growth Consultant

Outcome

Email response rate (old generic approach)2-3%

Email response rate (Role Prompting approach)28% (first 10 sends)

Meetings booked from this specific email3 out of 7 sent

Contract won (Rang Dhanu)৳1.5 lakh/month retainer — 3-month deal 🔥

Key InsightRole dile Claude automatically: specific language of that profession use করে, relevant concerns address করে, appropriate tone adopt করে, field-specific best practices apply করে। এটা একটা shortcut to expertise।

2

Technique 02

CHAIN-OF-THOUGHT PROMPTING

Complex problems-এ Claude-কে বলো step-by-step think করতে। "Think through this carefully" বা "Walk me through your reasoning" — এই phrases Claude-এর reasoning quality dramatically উন্নত করে।

Real Scenario — Sabbir (MBA Student, IBA Dhaka, Case Study Competition)

Business Case Analysis — Chain-of-Thought দিয়ে Competition জেতা

সাব্বির একটা national MBA case competition-এ participate করছে। Case: একটা Bangladeshi pharmaceutical company-র export strategy। Claude-এর normal answer ছিল surface-level। Chain-of-thought prompt দেওয়ার পর জিনিস পাল্টে গেলো।

তুমি একজন expert business strategy consultant, Harvard Business School থেকে MBA, McKinsey-তে ১২ বছরের experience, বিশেষত pharma industry এবং emerging market export strategy-তে। এই business case analyze করো। Step-by-step thinking দরকার — conclusion-এ jump করো না। Case: "MediCare BD" — Dhaka-based generic pharmaceutical manufacturer। Current: ৳450 crore annual revenue, 95% domestic sales। Goal: Export to 3 new markets in 2 years। Potential markets: Nepal, Myanmar, Vietnam। Constraints: Limited export experience, WHO-GMP certification pending (6 months), capital: ৳50 crore available। আমাকে চাই: Step 1: প্রতিটা market individually analyze করো — market size, regulatory barriers, competition, cultural fit Step 2: Our competitive advantage assessment করো honestly Step 3: Risks rank করো — probability × impact matrix Step 4: A final recommendation with reasoning — which market(s) first and why Step 5: First 90-day action plan with specific milestones "I recommend X" — এই shortcut নয়। প্রতিটা step-এর reasoning দেখাও।

Step 1: Market-by-Market Analysis

Nepal:
Market size: ~$180M pharmaceutical market, growing 12% YoY. Regulatory: Nepal Pharmacy Council — relatively straightforward for generic drugs, no WHO-GMP mandatory for most generics (our 6-month delay becomes non-issue here). Competition: Indian generics dominate (70%+ share) — but pricing advantage for Bangladeshi manufacturers is 15-25% cheaper due to lower labor costs. Cultural fit: Strong — similar disease burden patterns (diabetes, hypertension, respiratory). Bangladesh already exports ৳120 crore of goods to Nepal annually — existing trade relationships. VERDICT: Most accessible entry point.

Myanmar:
Market size: ~$220M, but politically unstable since 2021 coup. Payment risk: banking sanctions complicate USD transactions. Regulatory: FDA Myanmar — functional but inconsistent enforcement post-coup. Competition: Indian, Thai, Chinese generics — but less consolidated than Nepal. Currency risk: Kyat depreciation 50%+ since 2021. VERDICT: High potential, but timing is wrong. Revisit 18-24 months.

Vietnam:
Market size: ~$7.7B — significantly larger. Regulatory: Ministry of Health, Vietnam Drug Administration — requires local partner, registration takes 18-24 months minimum. WHO-GMP mandatory for import. Competition: Intense — European, Indian, Japanese. Capital required for market entry: ৳30-40 crore minimum. VERDICT: Too capital-intensive and complex for 2-year timeline given our constraints.

Step 2: Our Competitive Advantage — Honest Assessment
Strengths: Price competitiveness (real, 15-25% vs India), existing DGDA regulatory capability, proven domestic manufacturing quality. Weaknesses: WHO-GMP pending (eliminates Vietnam immediately), zero export infrastructure, no international regulatory experience, limited brand recognition. One honest insight: Our advantage is PRICE, not quality differentiation. This works in Nepal but would fail in Vietnam against European competition.

Step 3: Risk Matrix
Regulatory delay (WHO-GMP extends beyond 6 months): Probability 35%, Impact High — MITIGATE by starting Nepal now (doesn't need it). FX risk in Myanmar: Probability 70%, Impact High — AVOID for now. Key personnel loss (export team): Probability 40%, Impact Medium — hire 2 experienced export managers before Day 1.

Step 4: Final Recommendation
Enter Nepal in Month 1-6. Myanmar in Month 18-24 (political reassessment checkpoint). Vietnam in Year 3-4 (after WHO-GMP and Nepal learning curve). Reasoning: Nepal is the only market that matches all three of our constraints simultaneously — capital, timeline, and regulatory readiness.

Step 5: First 90-Day Action Plan
Days 1-30: Hire export manager with Nepal experience, engage Kathmandu-based distributor (3 candidates identified: XYZ, ABC, DEF), initiate Nepal regulatory dossier preparation.
Days 31-60: Distributor agreement signed, first shipment of 5 core products (oral solids — lowest regulatory barrier), regulatory submission to Nepal Pharmacy Council.
Days 61-90: First commercial sales in Kathmandu Valley, feedback loop established, WHO-GMP audit preparation starts in parallel.

When to Use Chain-of-ThoughtComplex decisions, business analysis, technical architecture, legal interpretation, financial modeling — যেকোনো জায়গায় intermediate reasoning steps গুরুত্বপূর্ণ। Simple tasks-এ এটা overkill — unnecessary হয়। Rule: যদি তুমি নিজে paper-এ steps লিখে solve করতে, Claude-কেও সেভাবে বলো।

3

Technique 03

FEW-SHOT EXAMPLE PROMPTING

Claude-কে examples দাও — "এই style-এ লিখো।" একটা ভালো example ১০০০ words of instruction-এর চেয়ে ভালো কাজ করে। Consistent brand voice, specific tone, particular format — সব examples-এ capture হয়।

Scenario — Fatema (Social Media Manager, Cosmetics Brand)

Brand Voice Consistency-র Problem

ফাতেমার brand "Glow BD" — playful, Gen-Z, Bangla-heavy, emoji-rich। Claude বারবার formal English দিতো। Few-shot fix করলো।

এই style-এ Instagram captions লেখো: Example 1: "নতুন serum আসছে 🌟 তোমার skin-এর জন্য যেন একটা vacation! ৩ দিনেই দেখবে পার্থক্য। কমেন্টে বলো — তুমি কোন skin concern নিয়ে বেশি worry করো? 👇 #GlowBD #কোরিয়ান_স্কিনকেয়ার" Example 2: "Sunscreen না লাগালে সব skincare waste! ☀️ এই summer-এ আমরা আনলাম SPF50+ — হালকা, sticky না, আর বাংলাদেশের গরমে perfect। ৳750-তে পাচ্ছো। Link bio-তে! 💛" এখন এই product-এর জন্য caption লিখো: Product: নতুন Vitamin C serum Key benefit: 7 দিনে brightness বাড়ে Price: ৳1,200 CTA: DM করতে বলো

Vitamin C magic শুরু হোক আজ থেকেই! ✨

7 দিনে glowing skin? হ্যাঁ, সত্যি বলছি! 🌟 আমাদের নতুন Vitamin C Serum তোমার face-এর dullness দূর করবে আর brightness আনবে — যেন ভেতর থেকে light জ্বলছে!

✨ 7 দিনে visible difference
🍊 Pure Vitamin C formula
💧 Oil-free, lightweight

মাত্র ৳1,200-তে পাচ্ছো। DM করো এখনই — stock limited! 👇

তোমার skin এখন কেমন? Comments-এ বলো! 💬
#GlowBD #VitaminCSerum #SkincareRoutine #GlowingSkintips #বাংলাদেশে_স্কিনকেয়ার

FEW-SHOT কোথায় কাজ করে

• Brand voice: Company-র specific communication style
• Legal writing: Specific contract language patterns
• Technical docs: Specific documentation format
• Email templates: Specific company communication style
• Social media: Platform-specific content style
• Code style: Team's coding conventions

✓ HOW MANY EXAMPLES?

• 1 example = basic style guidance
• 2-3 examples = strong pattern learning
• 4+ examples = very consistent output
• Too many = prompt becomes too long
• Sweet spot: 2-3 diverse examples

4

Technique 04

CONSTRAINT PROMPTING

Claude-কে বলো কী করবে না। "Avoid X," "Don't use Y," "No Z format" — negative constraints output-কে dramatically focused করে। Claude by default verbose। Constraints সেটা কাটে।

WITHOUT CONSTRAINTS: "আমার startup pitch deck-এর executive summary লিখো।" → Claude দেবে 600 words → Generic phrases: "We are excited to..." → No specific numbers → Multiple sections you didn't ask for

WITH CONSTRAINTS: "আমার startup pitch deck-এর executive summary লিখো। CONSTRAINTS: - Maximum 150 words, not a word more - No generic opener ("We are excited/pleased...") - Start with the PROBLEM, not the company - Include 3 specific numbers (market size, traction, ask) - No buzzwords: "innovative", "disruptive", "game-changing" - End with a single clear ask"

Real Scenario — Imran (Startup Founder, AgriTech, Sylhet)

Investor Pitch Deck Executive Summary — Constraint Prompting দিয়ে

ইমরান একটা AgriTech startup-এর founder। Investor presentation-এর জন্য executive summary লিখতে হবে। ৫ বার try করেছে — বারবার too long, too generic। Constraint prompting-এ first try-তেই হলো।

তুমি একজন expert startup pitch writer যে Y Combinator এবং Seedstars World-এর winning decks লিখেছো। আমার startup: "HarvestAI BD" Product: AI-powered crop disease detection via smartphone camera for Bangladeshi farmers Market: 16.5 million small farmers in Bangladesh Traction: 3,200 farmers onboarded in 4 months across Sylhet division Revenue model: ৳99/month subscription + agri input partnership referral fees Ask: $500,000 seed round Executive summary লিখো। HARD CONSTRAINTS: - Exactly 120-150 words (count করো) - শুরু করো problem statement দিয়ে — company name দিয়ে না - এই words use করো না: innovative, cutting-edge, revolutionary, game-changing - Specific data points: 3টা number minimum - End with exact ask and use-of-funds (one line) - Language: English (investor-facing) - No passive voice

Bangladesh loses ৳4,200 crore annually to preventable crop disease — largely because 16.5 million small-scale farmers lack access to timely agricultural expertise. They currently rely on word-of-mouth and delayed government extension services, often diagnosing disease-at-harvest when intervention is no longer possible.

HarvestAI BD solves this with a smartphone-based crop disease detection system that delivers real-time diagnosis and treatment recommendations in Bengali, offline-first, designed for feature phones.

In four months, 3,200 farmers across Sylhet division have onboarded at ৳99/month, validating both willingness-to-pay and adoption in low-digital-literacy populations. Our agrochemical referral partnerships already generate 30% of revenue.

We are raising $500,000 to expand to Rajshahi and Dinajpur divisions, targeting 25,000 active subscribers by Q4 2025. Funds will deploy across product engineering (40%), farmer acquisition (45%), and regulatory compliance (15%).

Word count: 148 words ✓ | Numbers: 5 ✓ | Zero buzzwords ✓

Before constraint prompting (5 attempts)All unusable — too long, too generic

With constraint prompting (1 attempt)Investor-ready — submitted to Seedstars BD event ✓

ResultShortlisted for final 10 teams 🔥

Master Constraint List — Copy করো"Maximum [N] words" · "No passive voice" · "No filler phrases" · "Start with [X] not [Y]" · "Bullet points only, no prose" · "Include exactly [N] examples" · "No academic language" · "Avoid these words: [list]" · "End with a single question" · "Format: [specific format]"

5

Technique 05

ITERATIVE REFINEMENT

Perfect output প্রথম prompt-এই expect করো না। Professional users iterate — ৩ step-এ perfect output পাওয়া যায়। প্রথম draft পেলে refine করো, তারপর polish করো।

1

GENERATE — First Draft

Full context দিয়ে initial output পাও। এই stage-এ perfect expect করো না — direction দেখো। Claude কী বুঝেছে, কী miss করেছে সেটা দেখো।

Prompt example: "একটা [TYPE] লিখো। Context: [CONTEXT]।"

2

REFINE — Fix What's Wrong

Same conversation-এ specific feedback দাও। "এটা too formal করো" না — "Third paragraph টা too defensive শোনাচ্ছে। এটা confident কিন্তু humble করো। আর opening line বদলাও — এখন too generic।"

Key: Specific feedback দাও, "ভালো করো" না।

3

POLISH — Final Touch

"এখন পুরোটা একটু tighten করো — ১০% shorter, আরো punchy। আর আমার name [NAME] add করো ending-এ।"

এই ৩-step process-এ ৯৫% output publication-ready হয়। Most users ১ step-এ ছেড়ে দেয় এবং বলে "AI কাজের না।"

Real Scenario — Rashid (VP Marketing, FMCG Company)

Press Release — ৩-Step Iteration দিয়ে Publication-Ready

রাশিদ ভাই একটা new product launch-এর press release লিখছেন। Media-র জন্য। প্রথম attempt ভালো ছিল কিন্তু perfect না। ৩ iteration-এ DS & STAR-ready হলো।

ITERATION 1: "আমাদের new energy drink 'Surjo' launch-এর জন্য press release লিখো। Bangladeshi market। ৳45/can। Target: urban youth।" → Claude দিলো decent first draft। কিন্তু: too corporate, lacked local flavor, no quotes

ITERATION 2: "ভালো শুরু। কিন্তু: 1. Quote দরকার CEO এবং একজন এবং youth ambassador-এর 2. Dhaka launch event detail add করো (Bashundhara City, July 15) 3. Opening পুরো বদলাও — এখন too generic corporate 4. Local energy drink market context add করো"

ITERATION 3: "Almost there। Final polish: - 10% shorter overall - Headline আরো punchy করো - Boilerplate section add করো (company bio) - Media contact placeholder দাও - AP Style date format করো"

6

Technique 06

AUDIENCE-FIRST PROMPTING

Claude-কে target audience describe করো — যাদের জন্য content, তাদের background, knowledge level, concerns, language preference। Claude automatically adjust করে।

Same Content, Different Audiences

Diabetes Management Explanation — ৩ Different Audiences

For Patient (Class 5 education level): "Explain type 2 diabetes management. Audience: 55-year-old Bangladeshi farmer, class 5 education, Bangla only, no medical terms." For Medical Student: "Explain T2DM management algorithms. Audience: 3rd year MBBS student, focus on HbA1c targets, drug selection criteria, when to escalate." For Hospital Board: "Present T2DM management cost-effectiveness. Audience: hospital administrators, focus on readmission rates, resource allocation, ROI of education programs."

AUDIENCE VARIABLES TO SPECIFY

📍 Education level
📍 Professional background
📍 Language preference
📍 Knowledge of the topic
📍 Main concern/pain point
📍 Decision they need to make
📍 Emotional state (stressed? curious?)
📍 Time available (will they read 500 words?)
📍 Cultural context (Bangladesh-specific?)

7

Technique 07

TEMPLATE CHAINING — WORKFLOW AUTOMATION

একটা Claude output পরের prompt-এর input হয়। Chain করে পুরো workflow automate করো। Human judgment টা strategic points-এ রাখো, repetitive execution টা Claude করুক।

Real Scenario — Shireen (Content Agency Owner, 3-person team)

Complete Blog Post Workflow — 4-Step Chain

শিরিন একটা content agency চালায়। প্রতি blog post-এ আগে ৬-৭ ঘণ্টা লাগতো। Template chaining দিয়ে ২-৩ ঘণ্টায় নামিয়ে এনেছে।

CHAIN STEP 1 — Research & Outline: "Topic: [TOPIC]। Audience: [AUDIENCE]। একটা comprehensive outline দাও। Include: 8 H2 headings, 3-4 bullets প্রতিটায়, word count suggestion per section, 3 key data points to research, 1 expert quote placeholder per section। NO prose yet — structure only।" → Output: Full article skeleton

CHAIN STEP 2 — Write Section by Section: "এই outline এর [SECTION NAME] টা লিখো। [PASTE THAT SECTION'S BULLETS] Guidelines: - 300-400 words - Conversational but authoritative - Real Bangladesh example include করো - Data point: [INSERT RESEARCH DATA]" → Output: One polished section

CHAIN STEP 3 — SEO Optimization: "এই article-টা দেখো: [PASTE DRAFT] Primary keyword: [KEYWORD] Secondary keywords: [LIST] Optimize করো: - Title + meta description (155 chars) - First 100 words-এ keyword naturally - H2 headings-এ keyword variation - Internal link placeholders: [INTERNAL LINK: topic] - Alt text suggestions for images"

CHAIN STEP 4 — Repurpose: "এই article থেকে তৈরি করো: 1. LinkedIn post (300 words, professional) 2. Twitter thread (8 tweets, numbered) 3. Instagram carousel (7 slides, each 2 lines) 4. Email newsletter (200 words, CTA included) 5. WhatsApp broadcast (100 words, casual) সব একসাথে।"

Agency Workflow ROI

Old workflow (manual): Blog post + 5 repurposed pieces6-7 hours/blog

New workflow (4-step chain)2.5-3 hours/blog (including review)

Monthly output capacity (3-person team)12 blogs → 24 blogs/month

Additional revenue potential12 extra blogs × ৳15,000/blog = ৳1,80,000 extra/month 🔥

Critical Mistakes

৮টা Mistake যা সবাই করে
— এবং কীভাবে Fix করবে

এই mistakes-গুলো থেকে বাঁচলেই তুমি ৮০% user-এর চেয়ে ভালো prompts লিখবে।

01

VAGUE TASK

✗ "Summary দাও"
✓ "এই 5-page report-এর 3-bullet executive summary দাও — decision-makers-এর জন্য, ২০০ words max।"

02

NO CONTEXT

✗ "Email লিখো"
✓ Who is sending? To whom? What's the goal? What's the relationship history? What's the tone?

03

ONE ATTEMPT

✗ "Output ভালো না" → Give up
✓ "Third paragraph টা too formal। আরো conversational করো। Opening বদলাও।" → Iterate.

04

NO FORMAT SPEC

✗ Claude দেবে যা মনে চায় — paragraphs, bullets, headers mix
✓ "Bullet points only, 3 per section, each max 2 lines।"

05

TOPIC SWITCH MID-CHAT

✗ Same conversation-এ unrelated topics mix করা — context dilutes
✓ Different project = new conversation। একটা chat = একটা focused project।

06

BLIND TRUST

✗ Claude-এর facts verify না করে publish করা
✓ Specific facts, statistics, citations — সবসময় independent source দিয়ে verify করো।

07

NO NEGATIVE CONSTRAINTS

✗ শুধু বলো "কী চাই" — "কী চাই না" বলো না
✓ "No buzzwords, no passive voice, no generic openers, max 150 words।"

08

SENSITIVE DATA

✗ Real customer names, NID, financial data paste করা
✓ Always anonymize — "[CLIENT NAME]", "[AMOUNT]", "[LOCATION]" placeholders use করো।

Deep Dive

Real Life Scenarios
— Before vs After

এবার দেখো একই মানুষ prompt technique শেখার আগে এবং পরে কীভাবে different results পেয়েছে।

Scenario 01 — Taslima (Secondary School Teacher, Comilla)

HSC Chemistry — Difficult Concept Explanation for Struggling Students

তাসলিমা ম্যামের class-এর students Electrochemistry বুঝছে না। Galvanic cell, electrode potential — এগুলো abstract। তিনি Claude দিয়ে explanation চাইলেন।

❌ Before (Weak Prompt)

"Galvanic cell explain করো।"

Output: Standard textbook definition। Students already read this and didn't understand.

No local context। No analogies। No misconception addressing।

✓ After (Engineered Prompt)

"তুমি একজন exceptional secondary school Chemistry teacher যে abstract concepts-কে everyday Bangladesh life দিয়ে explain করো। আমার Class 11-12 students galvanic cell বুঝছে না — specifically কেন spontaneous reaction হয় এবং electron flow-এর direction। Explain করো: 1. একটা local Bangladesh analogy দিয়ে (rickshaw, ferry, বাজার — যেটা fit করে) 2. Step-by-step — কোথায় কী হচ্ছে visual description 3. Students-এর সবচেয়ে common misconception address করো 4. ৩টা MCQ দাও (easy, medium, tricky) with explanations Language: Bangla-English mix (classroom style)।"

Galvanic Cell — দোকানদার analogy দিয়ে বুঝি

কল্পনা করো Dhaka-র একটা পুরনো বাজারে দুটো দোকান পাশাপাশি।

দোকান A (Zinc electrode/Anode): এই দোকানদারের কাছে অনেক "টাকা" (electrons) আছে। সে সবসময় খরচ করতে চায় — তাই এখানে oxidation হয়। Zn → Zn²⁺ + 2e⁻

দোকান B (Copper electrode/Cathode): এই দোকানদার টাকা নিতে পছন্দ করে — তাই reduction হয়। Cu²⁺ + 2e⁻ → Cu

Wire (External circuit): দুই দোকানের মধ্যে hundi — electrons এখান দিয়ে flow করে। A থেকে B — এটাই electric current।

Salt Bridge: দুটো দোকানের account balance maintain করার হিসাবরক্ষক। Ion flow করে, charge neutral রাখে।

Spontaneous কেন? কারণ Zinc স্বভাবতই electron দিতে চায় (higher reducing tendency), Copper নিতে চায় (Cu²⁺ reduction tendency higher)। Nature এই "চাওয়া"কে allow করে — তাই energy release হয়।

Common Misconception: "Current electron-এর direction-এ যায়।" — WRONG। Current conventional direction (B→A বাইরে), electron direction (A→B)। উল্টো! অনেক student এতে confuse হয়।

MCQ 1 (Easy): Galvanic cell-এ cathode-এ কোন process হয়?
(A) Oxidation (B) Reduction ✓ (C) Neutralization (D) Both
Answer: B — Cathode সবসময় reduction (remember: Red Cat = Reduction at Cathode)

Teacher Feedback

Class understanding before (show of hands)4 out of 35 students understood

After using this explanation next class28 out of 35 understood (showed hands)

Test scores (next week)Class average: 62% → 79% on electrochemistry questions 🎯

Scenario 02 — Mahbub (Entrepreneur, Food Delivery Startup, Sylhet)

Business Problem Solving — "Should I Expand or Fix First?"

মাহবুব ভাই-এর একটা food delivery startup আছে Sylhet-এ। দুটো option নিয়ে confused: Chittagong-এ expand করা নাকি Sylhet-এর existing operations fix করা। Chain-of-thought prompt দিয়ে decision নিলেন।

তুমি একজন expert startup advisor এবং operations consultant যে Bangladesh-এর food delivery market সম্পর্কে গভীরভাবে জানো। আমার situation: Company: "QuickEat Sylhet" — food delivery startup Current: Sylhet city, 180 active restaurants, 2,200 monthly orders Revenue: ৳8.5 lakh/month (GMV), take rate 18% = ৳1.53 lakh revenue Team: 12 people (4 delivery, 3 tech, 3 ops, 2 sales) Burn rate: ৳2.1 lakh/month Current problems: 34% on-time delivery rate (target: 80%), customer retention: 42% (target: 65%), app crash rate: 8% Option A: Fix Sylhet first — invest in delivery infrastructure, fix app, improve retention. Timeline: 4-6 months. Option B: Expand to Chittagong now — 1st mover advantage (no competitor there yet). Timeline: expand in 2 months. Step-by-step analyze করো এবং একটা clear recommendation দাও। আমার investor meeting আগামী সপ্তাহে।

My analysis — I'll be direct because you have an investor meeting.

Step 1: What the numbers actually say
You are burning ৳2.1L and earning ৳1.53L revenue — that's a ৳57,000/month cash deficit. Your unit economics are broken: 34% on-time delivery means 2 out of 3 orders arrive late. That 42% retention means 58% of customers who try you once never come back. These are not "growing pains" — these are fundamental product failures.

Step 2: What Chittagong expansion would actually mean
Expanding now would require approximately ৳8-12 lakh in new market costs (hiring, onboarding restaurants, marketing). With your current burn, you have roughly 4-5 months of runway if nothing changes. Chittagong would halve that — 2-3 months. You would be expanding a broken product into a new city with money you don't have, creating two broken operations instead of one.

Step 3: The 1st mover advantage argument
"1st mover" matters only if you can hold the market. With 34% on-time delivery, you would enter Chittagong, disappoint customers, get 1-star reviews, and create a permanent negative brand perception — giving any competitor who enters later 6 months from now a quality advantage. Bad 1st mover is worse than no 1st mover.

Step 4: My Recommendation — Fix First, Then Expand
4-month roadmap: Month 1-2: Fix app (crash rate to <1%), hire 1 experienced delivery operations manager (non-negotiable). Month 2-3: Pilot new delivery zone clustering in Sylhet — target 65% on-time. Month 4: Hit 65% on-time and 55% retention. Only then discuss Chittagong.

For Your Investor Meeting
Tell investors: "We are intentionally fixing unit economics before geographic expansion. Here are our 90-day KPI targets and our path to cash-flow positive in Sylhet before we ask for expansion capital." This shows strategic maturity. Investors respect this far more than premature growth stories.

Scenario 03 — Tanvir (Full-Stack Developer, Remote Job, Dhaka)

Complex Algorithm Problem — Job Interview Preparation

তানভির একটা US-based company-র remote job interview-এর জন্য prepare করছে। LeetCode-style coding problems। Chain-of-thought + constraint prompting combine করলো।

তুমি একজন expert competitive programmer এবং coding interview coach যে FAANG companies-এ কাজ করেছো। এই coding problem solve করো। কিন্তু answer directly দিও না — আমাকে শেখাও। Problem: "Given an array of integers, find two numbers that add up to a target sum. Return their indices. Assume exactly one solution exists. Cannot use same element twice." Example: nums=[2,7,11,15], target=9 → Output:[0,1] (because nums[0]+nums[1]=9) আমাকে দাও: 1. Problem বুঝতে সাহায্য করো — কী edge cases আছে? 2. Naive/brute force solution প্রথমে (O(n²)) — কেন কাজ করে বুঝিয়ে 3. Optimal solution (O(n)) — step-by-step intuition building 4. Why the optimal works — না মুখস্থ করিয়ে বুঝিয়ে 5. Python implementation — commented 6. Time + Space complexity analysis 7. Follow-up questions interviewer করতে পারে (3টা) CONSTRAINT: Code-এর আগে প্রতিটা approach-এর intuition plain Bangla-English-এ explain করো।

Python — Two Sum Solution (Claude-এর explanation সহ)

# ═══ APPROACH 1: BRUTE FORCE ═══════════════════════════════ # Intuition (Bangla): প্রতিটা number-এর সাথে বাকি সব number # এক এক করে add করো। যদি target পাই — return করো। # এটা দুটো nested loop — তাই O(n²)। def two_sum_brute(nums: list[int], target: int) -> list[int]: n = len(nums) for i in range(n): for j in range(i + 1, n): # j starts from i+1 (same element না) if nums[i] + nums[j] == target: return [i, j] return [] # Problem guarantees a solution, but good practice # Time: O(n²) — nested loops # Space: O(1) — no extra data structure # ═══ APPROACH 2: HASH MAP (OPTIMAL) ════════════════════════ # # Intuition (বাংলায়): # তুমি যখন nums[i] দেখছো, তুমি জানতে চাও: # "target - nums[i] কি আগে এসেছে?" # # যদি nums[i] = 7, target = 9, তাহলে তুমি খুঁজছো 2 (9-7=2)। # Hash map-এ আগের সব number store করো। # প্রতি step-এ O(1) lookup — তাই মোট O(n)। # # Analogy: বাজারে কেনাকাটা। তোমার কাছে list আছে। # প্রতিটা জিনিস দেখে মনে করো — "এর complement কি কিনেছি?" # পকেট check করার মতো — constant time। def two_sum_optimal(nums: list[int], target: int) -> list[int]: seen = {} # {number: index} — আগে দেখা numbers for i, num in enumerate(nums): complement = target - num # আমার জন্য কে দরকার? if complement in seen: # O(1) lookup return [seen[complement], i] # Found it! seen[num] = i # এই number record করো ভবিষ্যতের জন্য return [] # Time: O(n) — একবার array traverse # Space: O(n) — worst case সব number store হয় # ═══ TEST CASES ════════════════════════════════════════════ assert two_sum_optimal([2,7,11,15], 9) == [0,1] # Basic assert two_sum_optimal([3,2,4], 6) == [1,2] # Not adjacent assert two_sum_optimal([3,3], 6) == [0,1] # Duplicates print("All tests passed ✓") # ═══ FOLLOW-UP QUESTIONS INTERVIEWER করতে পারে ═══════════ # # Q1: "What if multiple solutions exist?" # → Return all pairs, not just first # # Q2: "What if numbers can repeat more than twice?" # → Store list of indices: seen[num] = [i1, i2...] # # Q3: "What if the array is sorted? Can you do better?" # → Two-pointer approach: O(n) time, O(1) space (even better!)

Scenario 04 — Roksana (Program Manager, Environmental NGO, Cox's Bazar)

Grant Proposal — UN Small Grants Programme Application

রক্সানা আপু একটা climate resilience project-এর জন্য UN grant-এ apply করবেন। Deadline কাল। Manual হলে ৪-৫ দিন লাগতো। Prompt engineering দিয়ে ১ দিনে submit করলেন।

তুমি একজন expert grant writer যে UN agencies, USAID, EU grants-এ ১৫+ বছর কাজ করেছো এবং $50M+ in successful grants write করেছো। আমার project: Organization: "Green Coast BD" — registered NGO, Cox's Bazar Project: Community-Based Mangrove Restoration and Climate Resilience Target area: Teknaf Peninsula coastal belt, 12 villages, 3,400 direct beneficiaries Duration: 18 months Ask: $75,000 USD Activities: 1. Mangrove restoration — 150 hectares, community-led 2. Alternative livelihoods (beekeeping, crab farming) — 200 households 3. Climate risk education and early warning system — 12 village committees 4. Women's leadership — 6 women-led forest management committees Grant: UN Environment Programme (UNEP) Small Grants Programme SDG alignment: SDG 13 (Climate), SDG 14 (Life Below Water), SDG 5 (Gender Equality) Project Narrative লিখো: Section: Problem Statement (600-700 words) Requirements: - Open with specific, compelling data (not generic "climate change is a problem") - Cox's Bazar specific — geography, vulnerability - Rohingya crisis environmental impact (relevant context) - Gender dimension — women disproportionately affected - Cite specific data: IPCC, Bangladesh government reports where applicable - No passive voice - UN grant language and tone - End with theory of change connecting problem to our solution CONSTRAINT: No generic opener. No "In recent years..." No "Climate change is a growing concern..."

PROBLEM STATEMENT

In 2023, Cyclone Mocha made landfall 60 kilometers from Teknaf Peninsula with 215 km/h sustained winds — the most powerful storm to strike Bangladesh in 11 years. Of the 3,400 households in the project's 12 target villages, 1,847 suffered structural damage. Recovery remains incomplete two years later. This is not exceptional — it is the new normal for Bangladesh's southeastern coast.

Cox's Bazar District, home to the world's largest refugee settlement and 2.6 million residents, sits at the intersection of three converging crises: accelerating climate change, catastrophic ecosystem degradation, and acute poverty. The district loses approximately 0.84 meters of coastline annually to erosion — a rate that has doubled since 1990 (BWDB, 2022). Sea surface temperatures in the Bay of Bengal have risen 0.5°C above the global average rate, intensifying cyclone intensity and frequency (IPCC AR6, 2021).

The immediate driver of coastal vulnerability is ecological collapse. The Teknaf Peninsula has lost 68% of its original mangrove cover over three decades — from 12,400 hectares in 1989 to 3,980 hectares in 2022 (Bangladesh Forest Department, 2023). Mangroves served as the primary natural barrier reducing storm surge by up to 66% and providing nursery habitat for coastal fisheries that sustain local livelihoods. Their loss is not a background condition — it is the mechanism through which climate change kills people and destroys economies in this specific geography.

The Rohingya displacement crisis since 2017 has compounded ecological stress in ways rarely acknowledged in climate discussions. The arrival of over 900,000 people in a 26-square-kilometer settlement required clearing approximately 6,500 acres of forested hillside for shelter, accelerating soil erosion into coastal waterways and contributing to sedimentation that has degraded an additional 1,200 hectares of coastal mangrove habitat adjacent to the project area.

Women and girls face disproportionate climate vulnerability in Teknaf — a dimension that standard vulnerability assessments consistently undercount. Social mobility restrictions mean women evacuate later and return earlier to flooded homes, increasing mortality risk. Post-disaster economic shocks fall disproportionately on households dependent on women's informal income — primarily small-scale aquaculture, homestead gardening, and collection of non-timber forest products — all of which are directly undermined by ecosystem degradation. Of the 200 target households in the livelihoods component, 78% are female-headed, following the death or permanent out-migration of male household members following previous cyclones.

Theory of Change: Community-led mangrove restoration, when connected to alternative livelihoods that depend on ecosystem health, creates aligned incentives for long-term stewardship. Coastal communities protecting mangroves they economically benefit from maintain restoration gains beyond project timelines. Women leading forest management committees institutionalizes climate resilience within community governance structures. This project does not introduce external solutions to passive communities — it removes barriers to agency that communities already demonstrate in informal practice.

Grant Writing ROI

Manual grant writing (problem statement alone)8-12 hours (research + draft + revise)

With prompt engineering approach90 minutes (prompt + 2 iterations + review)

Full grant narrative (all sections)1 day vs. 4-5 days

Application submittedOn time — shortlisted for review ✓ 🔥

Template Library

30+ Prompt Templates
Copy করো, Win করো

প্রতিটা template tested এবং real Bangladesh context-এ verified। Placeholder [CAPS]-এ — replace করো।

✍️ WRITING TEMPLATES

• Article: Role + Niche + Audience + SEO keyword + tone + word count
• Email: Role + relationship + goal + tone + constraints + length
• Press Release: Event + key messages + quote placeholders + AP style
• Social Caption: Example-based + platform + CTA + hashtag count
• Product Description: Benefits-first + audience + price anchor + CTA

💻 CODING TEMPLATES

• Debug: Language + Framework + Full error + Code + Root cause + Fix + Test
• Code Review: Paste code + Check list (security, perf, style) + Severity rating
• Architecture: Feature + Scale + Stack + Team size + Constraints + Trade-offs
• Documentation: Endpoint + Models + Auth + Examples (Py/JS/cURL)
• Algorithm: Problem + Brute force first + Optimal intuition + Complexity

💼 BUSINESS TEMPLATES

• Proposal: Context + Pain point + Solution + ROI + Timeline + CTA
• Meeting Minutes: Raw notes + Attendees + Decisions + Action items + Owners
• Status Report: RAG status + Risks + Decisions needed + Next steps
• Investor Pitch: Problem + Solution + Market + Traction + Ask (hard constraints)
• Grant Proposal: Problem + Evidence + Theory of Change + SDG alignment

👩‍💼 HR TEMPLATES

• Job Description: Role + Skills + Culture + Benefits + Inclusive language
• Offer Letter: Name + Role + Salary + Start + Benefits + Acceptance deadline
• Warning Letter: Incident + Policy ref + Impact + Improvement expected + Timeline
• PIP: 30-day goals + Metrics + Check-in schedule + Consequences
• Performance Review: Raw manager notes → SMART goals format

🔬 RESEARCH TEMPLATES

• Paper Summary: Audience level + Methodology focus + Clinical implications
• Competitor Analysis: 5 competitors + 6 dimensions + Market gap
• Market Research: Geography + Segment + Data sources + Actionable insights
• Literature Review: Topic + Timeframe + Academic vs practitioner sources
• Survey Analysis: Paste data + Key findings + Visualization suggestions

🎨 CREATIVE TEMPLATES

• YouTube Script: Hook + Pattern interrupt + 3 sections + CTA + Timestamps
• Podcast Outline: Topic + Guest bio + 8 questions + Segment timing
• Ad Copy: AIDA formula + Platform constraints + A/B variations
• Brand Story: Founding + Problem + Mission + Customer hero
• Content Calendar: Platform + Pillars + Audience + Seasonal hooks

THE MASTER TEMPLATE — যেকোনো task-এ কাজ করে

তুমি একজন [ROLE — expert, experienced, senior] [PROFESSION/SPECIALTY] যে [RELEVANT EXPERIENCE/CONTEXT]। আমার situation: [BACKGROUND CONTEXT — company, industry, size, relationship] Task: [SPECIFIC DELIVERABLE — verb + object + format] Requirements: - [REQUIREMENT 1 — positive: include X] - [REQUIREMENT 2 — positive: format/structure] - [REQUIREMENT 3 — negative: avoid Y] - [REQUIREMENT 4 — constraint: max Z words / specific tone] Example of desired style/tone: [PASTE EXAMPLE — optional but powerful] Audience: [WHO WILL READ THIS — education, background, concerns]

This single template structure = 90%+ use cases covered

Comparison

Prompt Technique Quick Reference

Technique	When to Use	Key Formula	Improvement
Role Prompting	Any professional writing, advice, analysis	"তুমি একজন expert [ROLE] যে [EXPERIENCE]।"	3-5x more specific
Chain-of-Thought	Complex decisions, analysis, math, strategy	"Step-by-step think করো। Conclusion-এ jump করো না।"	2-4x more accurate
Few-Shot Examples	Brand voice, specific style, format consistency	"Example 1: [X]। Example 2: [Y]। এখন: [TASK]।"	90%+ style match
Constraint Prompting	When output is too long, too generic, wrong format	"Max 150 words। No [X]. Start with [Y]. Avoid [Z]।"	First-try usability 5x
Iterative Refinement	High-stakes documents, creative work	Generate → "Fix [specific issue]" → "Final polish"	Publication quality
Audience Prompting	Content for non-expert audiences, teaching	"Audience: [specific description — education, background]"	2x comprehension
Template Chaining	Repetitive workflows, content pipelines	Output of Step N = Input of Step N+1	60-75% time saved

Honest Review

Prompt Engineering-এর Realities

✅ What Prompt Engineering Gives You

✓ Consistently better output — not dependent on luck
✓ Faster iteration — less back-and-forth
✓ Reusable templates — one-time investment
✓ Professional-grade output from a free tool
✓ Skill that transfers to all AI tools
✓ Competitive advantage over colleagues who don't know this
✓ Able to delegate more work to AI reliably
✓ Compound returns — better prompts = faster learning

⚠️ What Prompt Engineering Won't Fix

✗ Hallucination — Claude can still make facts up; verify important claims
✗ Knowledge cutoff — recent events not in training data
✗ Real-time data — stock prices, live news, current web
✗ Computation errors — math can be wrong; verify calculations
✗ Clinical/legal/financial final judgment — human expert always needed
✗ Your unique voice — AI can imitate but you must add personality
✗ Creative originality — AI remixes training data; truly novel ideas are yours
✗ Relationship context — Claude doesn't know your history with a person

Resources

References & Further Reading

Anthropic Prompt Guide

docs.anthropic.com/en/docs/build-with-claude/prompt-engineering — Official Claude prompt engineering documentation। Best practices, examples।

Claude Model Card

anthropic.com/model-card — Capabilities, limitations, safety properties। Know what Claude can and cannot do।

Anthropic Research Blog

anthropic.com/research — Constitutional AI, chain-of-thought papers, capability research। Academic depth।

LMSYS Chatbot Arena

lmsys.org — Independent human preference benchmarks। Claude vs GPT vs Gemini on real tasks।

Learn Prompting (Free)

learnprompting.org — Open-source prompt engineering guide। Excellent for deep technical understanding।

FreeLearning365.com

Complete 20-post Bangla AI tutorial series। Prompt library, real Bangladesh scenarios, profession-specific guides।

Hashtags — Click to Copy

#PromptEngineering#ClaudeAI#AIBangla#কৃত্রিম_বুদ্ধিমত্তা#AIForWork#BangladeshTech#BetterAIOutput#AISkills#ProductivityHack#DigitalBangladesh#FreeLearning365#AIটিউটোরিয়াল#LearnAI#SmartWork#DeveloperBD#ContentCreatorBD#FreelanceBangladesh#AIHacks#WorkSmarter#FutureSkills

Coming Next

POST

06

MAXIMIZE FREE PLAN

Zero Cost Strategy

Free plan থেকে Pro-level output। Smart batching, morning power hour, template libraries, free tool combinations — সব কিছু। ৳0 invest করে ৳2,200 Pro-এর ৮০% value বের করার complete guide।

৳0

Total Investment Required

FreeLearning365.com↑ শুরুতে যাও

FREELEARNING365

Bangla AI Tutorial Series · Post 5 of 20 · Prompt Engineering — 10X Better Output

AnthropicClaude AIPrompt DocsLearn PromptingFreeLearning365

© 2025 FreeLearning365.com · All rights reserved · Powered by Claude AI · Post 5 of 20