Class 5 · 5.4

YOLO build: AI digital twin ke sath portfolio site

Aik instruction, aik zabardast Next.js site. Doosri instruction, aik AI jo aap ke career ke baare mein sawalat ke jawabat deta hai. Phir hum wapas aate hain aur har line ko samajhte hain.

🎯 Maqsad: site + digital twin ship karna⏱ Parhein: 12 min🧰 Tools: Cursor · OpenRouter

Ab waqai YOLO karne ka waqt hai. Hum 5.2 ke usool jaan boojh kar todenge — koi agents.md nahi, raaste mein kam checking — kyun ke throwaway MVP ke liye yeh theek hai. Phir aakhir mein wapas aakar professional finish denge. Yahan sab optional hai aur choose-your-own-adventure: aapke nataij alag honge, aur yahi point hai.

01Project set up karein

Cursor (ya apna IDE) mein aik fresh window kholain aur aik naya khali folder banayein — naam site. Do chhoti files bohot zaroori hain, inhein tawajjah se banayein.

`.env` banayein

Right-click → New File, naam exactly .env. Neeche diye format mein apni OpenRouter key paste karke save karein. Spelling exact honi chahiye, warna kuch nahi chalega.

`.gitignore` banayein

Doosri nayi file, .gitignore, jisme .env likha ho taake secret kabhi repository mein na jaye. Repo se pehle bhi yeh achhi aadat hai.

Apna profile add karein

LinkedIn profile ko PDF banayein (Resources → Save to PDF) — ya text file mein paste karein, ya CV use karein — aur site folder mein LinkedIn.pdf ke naam se daalein. Site isi se banegi.

.env

OPENROUTER_API_KEY=sk-or-v1-your-key-here

.gitignore

.env

Seedha OpenAI use kar rahe hain?

Agar OpenAI key use karni ho to variable ka naam OPENAI_API_KEY rakhein aur prompt us hisaab se adjust karein. Jo bhi choose karein, naam exactly match hona chahiye.

02Cursor ko YOLO mode mein daalein

Settings kholain (Ctrl/Cmd+Shift+J) aur teen changes karein.

Agents → Usage summary

“Always” set karein

Dikhaata hai ke har model aapke plan ka kitna use kar raha hai, neeche.

Agents → Auto Run

“Run everything, un-sandboxed”

Yeh YOLO switch hai. Agar abhi comfortable nahi hain to step-by-step approval rakhein.

Models

GPT-5.2 Codex (High)

Strong, mehnga option — yahan quality ke liye. Jo bhi top model pasand ho woh chunein.

Un-sandboxed ka matlab un-sandboxed

“Run everything” agent ko bina poochhe commands chalane deta hai. YOLO ka point yahi hai, lekin risk asal hai — sirf tab enable karein jab aap dekh rahe hon, aur kabhi aisi machine ya folder par nahi jise aap change hone nahi de sakte.

03Prompt 1 — site build karein

Full YOLO par hum agents.md skip karke aik hi instruction bhejte hain. “auto” model selection band karein, GPT-5.2 Codex High choose karein, aur yeh bhejein:

prompt → GPT-5.2 Codex High

Please build me a professional website running locally.
My LinkedIn profile is in LinkedIn.pdf.
Make the website stunning. Enterprise meets edgy.
It should include about me, my career journey, and links
to a portfolio for the future. Iterate to make it as slick
and professional as possible, and let me know when complete.
Use Next.js.

Agent plan karta hai, web/ sub-folder banata hai, aur kaam karta hai. Root mein error aaye to uske around nikal jata hai. Agar page na khule to sab se aalsi-lekin-kaam-karne wala YOLO move yeh hai: error seedha chat mein paste karein, koi explanation nahi, aur kaho “make it work.”

Chalayein

cd web phir npm run dev, localhost:3000 kholain, aur aapko profile se bani “enterprise-meets-edgy” site milni chahiye — about-me, career journey, portfolio link, even working email link.

04Aage barhne se pehle checkpoint

Abhi snapshot save karein

YOLO aik step mein cheezein barbaad kar sakta hai. Sahi tareeqa Git hai — har achhe point par commit karein (aur web/ ke andar Next.js ki extra repo hata dein taake sirf aik ho). Gareeb-aadmi wala version: poora site folder duplicate karke copy restore point rakh lein.

05Prompt 2 — digital twin add karein

Ab AI wala hissa. OpenRouter ke Models page se model chunein — free search karne se free tier dikhta hai; GPT-OSS open-source models ke free aur saste paid versions hain. Exact model name copy karke prompt mein daalein.

prompt → GPT-5.2 Codex High

Please now add the ability to chat with an AI "digital twin"
that can answer questions about my career. Please use OpenRouter.
My OpenRouter API key is in the .env file in the project root.
Please use the model named <paste-model-name>.
Make the changes, make sure it works, and let me know when ready.

Dobara chalayein to digital-twin chat section milega jo waqai OpenRouter call karta hai. Perfect nahi hoga — thora jankiness umeed karein (button jo kuch na kare, page galat jagah jump kare). Yeh normal hai, aur agla step isi ke liye hai.

06Jab aap rut mein phans jayein

Kabhi aap wahi fix chaar-paanch baar maangte hain aur model miss karta rehta hai — colours galat, layout toota. Frustration badhti hai. Loop todne wali technique:

“Don’t do it that way. Remove what you did and build a completely different chat interface.” — naya attempt aksar stuck wale ko debug karne se asaan hota hai.

Yahan kai failed patches ke baad isi reframe ne chat UI theek ki. Kab piece restart karna hai aur kab nudge karna hai — yeh core vibe-coding skill hai.

07Samjhein, phir review karwayein

Tezi se banane ke liye rules tode; ab unki izzat is tarah karein ke hum code ke malik banain. Do prompts zyada tar kaam kar dete hain — aur doosra aik genuinely powerful move dikhata hai.

Beginner tutorial maangein

“Complete beginner ke liye comprehensive Markdown tutorial likho: technology ka summary, high-level walkthrough, samples ke sath detailed code review, aur improvement ke paanch self-review suggestions.” Drill karte rahein — “isay barhao, mazeed do” — jab tak waqai samajh na aa jaye.

Aik alag model se review karwayein

Cursor mein model Claude Opus 4.5 par switch karein aur kaho: “Is project ka comprehensive code review karo aur review.md mein likho, remedial actions ke sath — koi code change mat karo.” Doosra, alag-trained model taza nazar late hai.

Review wapas feed karein

Review Codex ko bhejein: “Yeh parho aur remedial actions implement karo, ya kaho agar disagree ho.” Aapne abhi do frontier models ko aik codebase par collaborate karwaya.

Multi-agent kaam ki jhalak: alag models, har aik puzzle ka tukra uthata hai.

Ho gaya — aur aap iske malik hain

Aapke paas digital twin wali portfolio site, use explain karne wala tutorial, aur independent code review hai. Agar abhi nahi pahuncha: sabr karein, hard simplify karein, pehle saadi site chalayein, phir pieces wapas add karein. Live chahiye? Model se Vercel jaisi jagah ke liye deploy checklist likhwa lein.

✓ Aham baatein

Prompt se pehle .env + .gitignore set karein aur profile add karein.
YOLO = Auto Run “run everything un-sandboxed” strong model ke sath — dekhein, aur sirf wahan jahan safe ho.
Lagatar checkpoint karein (Git, ya folder copy) taake galat step kabhi fatal na ho.
Rut mein phans gaye? Patch karne ki jagah bilkul alag approach maangein.
Loop zimmedari se band karein: samajhne ke liye tutorial, harden karne ke liye cross-model review — code apna rakhein.