How did I build and deploy 3 fully functional apps while still in college, before I even landed my first internship?
No professor ever showed me how to do anything in college. So I did it all by trial and error — and that's why I was able to create a "real" portfolio before my SDE internship even began.
At some point, probably after the 100th time someone told me to "build projects," I finally figured this out: coding and shipping are two entirely separate skill sets.
I learned this the hard way, creating three fully functioning applications using AI technology as a CSE student. One application I created during a 4-hour sprint session.
I didn't do it because I am some form of 10x engineer. I did it because I stopped thinking about AI tools as nothing more than an autocomplete feature. I started thinking about AI tools as part of a development team.
Below is the stack of tools that allowed me to go from a "school project" to resolving production bugs on multiple microservices receiving thousands of concurrent requests while working as an SDE intern.
Tools No One Teaches You in Class
ChatGPT has been around forever. However, almost everyone doesn't use the entire AI developer stack that allows people to move beyond simply asking an AI tool to write a function to actually deploying products.
- Claude / Codex / Cursor / Windsurf / Replit Agent / Lovable — Used for making high-level design decisions and refactoring code, rather than just typing faster.
- Model Context Protocol (MCP) — The protocol allowing AI tools to communicate directly with your database and other parts of your codebase, rather than just trying to guess what they should do based on a vacuum.
- n8n — Lets you wire MCP-connected AI tools into your existing workflows, so instead of manually looping an AI call into every step, the workflow triggers it for you and you focus on higher-level decisions.
- OpenAI API + Gemini API — Used in conjunction with each other, based on cost and reasoning trade-offs per task.
Proof It Isn't Just a Theory
I'm not writing about what "the internet" says. That's what happened to the stack I used to create my portfolio when I was a student.
1. An AI Cold Outreach Platform (Gmail + OpenAI) — PitchPilot
- Lowered email generation time by 70%.
- Increased automated follow-ups by 45%.
- Implemented tenant isolation using stateless JWTs.
- Designed MongoDB schemas for individual tenants.
2. A Hotel Analytics Platform — AzzuroIQ
A platform delivering real-time insights using LLMs for four properties, built in a single 4-hour sprint using contract-first REST APIs.
- Reduced integration errors by 40%.
- Processed real-time LLM insights for four distinct hotel properties.
3. A Full-Stack Prediction Market Platform — MarketPulse
- Shipped 50% ahead of the deadline.
- Boosted category discovery by 35%.
- Architected structural MongoDB schemas for eight complex relational entities.
- Implemented secure Firebase authentication.
All of these were done outside of a bootcamp curriculum. They were built because I treated AI tools as a force multiplier for architecture and iteration speed — not as a crutch to write boilerplate code.
This portfolio is what got me an internship at my current company, where I've since fixed more than 15 production bugs and shipped three core features across live microservices.
Actual Workflow — Steal This
1. Build Your Data Model Before You Build Your UI
Every one of my projects started with the data model. AI tools are dramatically better at helping you build on top of a solid schema than helping you invent one from scratch mid-build.
2. Use Context Like MCP
Don't use copy-paste prompting. When your AI tool can see your actual repository or API instead of a pasted snippet, your bug rate drops hard.
3. Create New Workflows
This is the biggest opportunity AI offers.
The boring 40% of tasks will be owned by AI. You'll have control over the more difficult (and less repetitive) 20% of tasks.
Issues related to authentication, tenant isolation, and data consistency — the things that break production — require your attention. AI creates drafts, but you need to review them.
4. Don't Automate Creativity
n8n managed the looping outreach in PitchPilot, so I was able to focus on developing the creative prompt engineering required.
5. Get Something Done Quickly
In MarketPulse, we shipped 50% earlier than expected. It wasn't due to faster development time; it was because we cut our scope and let the AI tool stack handle all the heavy lifting.
This Matters More Than Another DSA Problem
I completed over 1,000+ DSA problems across LeetCode, CodeChef, Codeforces, and HackerRank.
Those efforts were enough to get past the interview round. However, those skills did nothing to assist with creating projects that added value to my resume or helped me fix production bug issues on microservices at a rate of 12% faster API response times.
The skill that is underrated right now isn't "Can you use AI?" — everyone can prompt ChatGPT!
It's "Can you architect a workflow where all of your AI tools hand off to each other cleanly?"
Just like how you would design a microservice boundary.
That's the part that feels like insider knowledge. But it's not gatekept — it's just not taught because most curricula are still catching up with how fast this stack is moving.
That's the pattern I keep coming back to:
Cut the noise. Let AI handle the repeatable stuff. Spend the saved time where it actually counts.
Anyway — if you made it this far, you're clearly the kind of person who reads past the headline, and I respect that.
Come say hi on LinkedIn, poke through the code on GitHub, or have a look at my Instagram — whichever feels less like a LinkedIn cold DM and more like a "hey, cool, same."
About Me
Graduated with a B.Tech in Computer Science & Engineering from IIIT Bhagalpur in 2026 and recently completed a stint as a Software Development Engineer Intern (SDE Intern).
During my internship, I worked primarily with:
- C#
- .NET Core
- Microservices
- MVC
- Azure DevOps
- SonarQube
I also continue to develop AI-native tools independently, exploring ways to combine AI, automation, and software engineering to build products that solve real problems.
