DEV Community: DipCodes

I tried running Qwen 2.5 Coder locally and accidentally built a full MERN stack app in 1 hour

DipCodes — Sat, 14 Mar 2026 17:40:20 +0000

Was trying to run Qwen 2.5 Coder offline on my laptop but ended up
building this full MERN stack project management app with 2FA using
Haiku 4.5 in one evening.

Here's exactly what I built and how:

What I built

A full ProjectHub app with:

Beautiful glassmorphism dark UI
JWT Authentication with 2FA (Google Authenticator)
Dashboard with project stats
MERN stack (MongoDB, Express, React, Node.js)
Vite + Tailwind CSS frontend

My setup (all free, ₹0)

Ollama — runs AI models locally on my laptop
Qwen 2.5 Coder 14B — local offline coding model
Claude Haiku 4.5 — via GitHub Copilot free access
Continue.dev — free Copilot alternative in VS Code

How it happened

Started by installing Ollama to run Qwen locally
Set up Continue.dev in VS Code
Haiku 4.5 auto-detected through GitHub Copilot
Gave it one detailed prompt for the full MERN stack
It created 20+ files automatically in minutes
Fixed static buttons and connected the backend
Added 2FA with Google Authenticator

The prompt that started everything

Build a full MERN stack project management app with:
- JWT auth + 2FA
- Glassmorphism dark UI
- Dashboard with stats
- Framer Motion animations
- Tailwind CSS
Create ALL files with complete working code.

Time taken

⏰ Total: ~2.5 hours from zero to working app with 2FA

Key takeaway

AI doesn't replace developers — it makes you 10x faster.
I still had to:

Know what to ask for
Fix bugs and connect everything
Understand the code it wrote
Make design decisions

The AI typed fast. I built the product.

Claude rocks anyways 🔥

claudeai #ai #mern #webdev #buildinpublic

Day 0 — I just committed to the most important 18 months of my life

DipCodes — Fri, 13 Mar 2026 18:06:39 +0000

I'm a first year engineering student in India.

Self taught. Building in public. 7 live projects already shipped.

Today I made a decision that scares me a little.

I'm giving myself 18 months to reach a goal that most people
would call unrealistic for someone in my position.

I'm not going to tell you what the goal is yet.

Let the work speak.

Here's what you'll see from me:

→ Daily building updates
→ Weekly project drops
→ Raw honest documentation of the journey
→ No tutorials. No fluff. Just real work.

Follow if you want to watch someone
go all in on something that matters.

Day 1 starts tomorrow.

— DipCodes

I built a tool that automatically tracks every free AI API daily — here's how

DipCodes — Thu, 12 Mar 2026 12:19:38 +0000

Every time I wanted to build something with AI, I wasted hours on the same frustrating loop.
Which free APIs actually work? What are the real limits? Do they need a credit card? The documentation lies. The Reddit posts are outdated. You only find out the truth after you've already integrated it.
So I built FreeStack AI to solve this for myself — and every other developer stuck in that loop.
What it does
FreeStack AI tracks 5 free AI APIs in real time — Groq, Mistral, Cohere, Gemini, and OpenRouter. Every 24 hours it automatically tests each one with real API calls, measures response speed, checks availability, and saves the results to a database.
No more guessing. You land on the site and instantly see what's working right now.
How the automation works
The core is a simple node-cron job running on a Node.js backend hosted on Render:
javascriptcron.schedule('0 0 * * *', async () => {
await testAllAPIs();
await saveResultsToDB();
});
Every night at midnight it hits all 5 APIs with real test calls, measures latency, and stores everything in MongoDB. The frontend on Vercel just reads from the database — so the data is always fresh without any manual work.
The hardest part — CORS in production
This one took me an entire day to debug.
Everything worked perfectly locally. The moment I deployed to Vercel + Render, requests were completely blocked. Classic CORS issue between two different production domains.
The fix was straightforward once I understood it — explicitly whitelisting the Vercel frontend URL in the Render backend:
javascriptapp.use(cors({
origin: 'https://freestack-ai.vercel.app'
}));
Local development hides this because everything runs on localhost. Production exposes it immediately. Lesson learned.
What's next
Right now 5 APIs are tracked. I'm expanding to 20+ soon — Together AI, Hugging Face, Replicate, DeepInfra and more.
There's also a code generator built in — select your use case, pick an API, choose your language, and get production ready starter code instantly.
Try it here — freestack-ai.vercel.app
If you've ever wasted time figuring out which free AI APIs actually work, this is built for you.