Lesson · Hinglish · ⏱ 8 min padhne mein

✍️ Prompting: LLM ko saaf instructions dena

Q: Aapki app ko LLM ka jawaab code mein padhna hai. Prompt mein kya maangna chahiye?

Ek specific JSON format. Fixed JSON shape reliably parse ho sakta hai. Free text nahi.

🏠 Asli zindagi ki misaal

Naye intern ko bolo 'report bana do' to kuch bhi milega. Bolo 'March ki one-page sales report banao, February se compare karo, table use karo, aur top 3 cities highlight karo', to bilkul wahi milega jo chahiye. LLMs brilliant interns hain: brief jitna saaf, kaam utna behtar.

💡 Aasaan shabdon mein

Acche prompt mein hota hai: role aur context, exact task, constraints jaise length, tone, language aur format, aur acche output ke examples. System prompt hamesha ke rules set karta hai ('Aap ek helpful bank assistant ho. Account numbers kabhi share mat karo.'). Few-shot prompting matlab input → chahiye output ke 2–3 examples dikhaana. Model se jawaab se pehle step by step sochne ko kehna reasoning behtar karta hai. Structured output (JSON) maangne se jawaab code mein use ho sakte hain. Prompts ko code ki tarah treat karo: real examples pe test karo aur improve karte raho.

🪜 Step by step

Role aur goal batao: "You are a support assistant for a bank. Answer customer questions about cards."
Model ko jo context chahiye do: policy text, customer ka plan, aaj ki date.
Rules aur limits saaf likho: sirf policy se jawaab do, warna "I don't know" kaho, 100 shabdon se kam rakho.
Acche jawaab ke ek do examples dikhao, aur output format define karo (bullets, naam wale fields ke saath JSON).
20 ya zyaada asli inputs pe test karo, failures dekho, ek baar mein ek cheez badlo, aur prompt ke versions rakho.

🔑 Zaroori shabd

System prompt: Hamesha ke instructions jo har response ko shape karte hain.
Few-shot: Prompt mein kuch solved examples daalna.
Structured output: JSON jaisa fixed format maangna taaki code use padh sake.

👀 Example dekho

text

System: You classify customer messages for a food app. Reply only with JSON.
User:   Classify: "Order came cold and 40 mins late 😡"
        Format: {"category": "delivery|food|payment|other", "sentiment": "pos|neg|neutral", "urgent": true|false}
Model:  {"category": "delivery", "sentiment": "neg", "urgent": true}

Role + task + exact format = reliable, machine-readable jawaab.

⚠️ Aam galtiyaan jinse bachna hai

"Behtar banao" jaise vague prompts. Batao behtar ka matlab kya: chhota, friendly, call to action ke saath.
Ek do try mein prompt judge karna. Ek example pe badhiya dikhne wale prompts aksar doosron pe fail hote hain.
Company se approved na hon aise tools ko bheje prompts mein secrets ya private data daalna.

✅ Jaldi se check karo

Aapki app ko LLM ka jawaab code mein padhna hai. Prompt mein kya maangna chahiye?

Ek friendly paragraph
Jitni ho sake utni detail
Ek specific JSON format

Jawaab dekho

C. Ek specific JSON format: Fixed JSON shape reliably parse ho sakta hai. Free text nahi.

🙋 Seekhne wale yeh poochhte hain

Kya lambe prompts behtar kaam karte hain?

Saaf, lambe se behtar hai. Jo zaroori hai woh daalo, jo nahi woh hatao. Lamba bekaar context model ko confuse kar sakta hai aur mehenga bhi padta hai.

Chain of thought kya hai?

Model se jawaab se pehle step by step sochne ko kehna. Maths aur multi step problems mein madad karta hai; kai naye models yeh andar hi karte hain.

📌 Yaad rakho

Role + context + task + format + examples. Phir test aur improve karo.

📖 Is lesson ke shabd (AI dictionary)

Poora course, aapke roadmap ke hisaab se

Progress tracking, quick checks aur step-by-step order ke saath. Yeh lesson in courses ka hissa hai:

AI Foundations Path · AI Product Engineer · LLM / Agent Engineer · AI Forward Deployed Engineer · AI Product Manager

Course mein seekho