When I realized my friend's resume was not even reaching recruiters and being rejected at first step by ATS, I started helping him through whatever I knew.

But then I saw a pattern. It wasn't just him. When I reached out to people I knew, many of them had the same story: "I have the skills, but can't get till the interview."

Phase 1: The playground

I didn't start by writing code or building the final tool. I started by playing around with what was available. Over the past few years, I’ve been exploring various AI tools both professionally and personally. So, naturally, my experiments began with the heavy hitters: ChatGPT and Google Gemini gradually moving to other tools.

(If you are new to the AI space and aren't familiar with these tools, here are some great resources to get you up to speed):

I started with a single prompt + Input Resume -> Output Feedback. It worked great for 1-on-1 interactions. But it lacked scale. How could I do this for thousands of people without manually prompting the AI for every single person?

Phase 2: The "Gem" Era

That’s when I realized I might already have the solution. I was using Gemini Gems at work to speed up my design workflows. If you’re new to Gems, think of them as custom versions of Gemini that "remember" your specific instructions so you don't have to repeat them every time.

(Curious about Gems? Check these out):

I thought I had cracked it. I created a "Resume Improver Gem." It did the task, and it was shareable. But as I tested it further, I hit a wall.

Phase 3: The refinement 'Hitting the wall'

While Gems are fantastic for personal productivity, I ran into specific limitations when trying to build a public solution:

• Accessibility: I wanted a tool anyone could use, even if they didn't have a Google account or a specific subscription.
• Interface: I needed a simple UI (upload button), not a chat window where users have to type prompts.
• Logic: I needed to handle moderately complex logic (like comparing a resume against a job description) which gets messy in a simple chat bot.
• Scalability: I needed something I could iterate on and improve day by day.

Phase 4: Taking the red pill

That is when I discovered tools like Google AI Studio and Emergent.

Now, I won't claim I was doing "Vibe Coding" (the trendy term for coding with AI vibes). I was just trying to solve a problem efficiently. I didn't jump straight to building a full web app. Instead, I started by creating small tests to understand the limitations of these platforms.

These experiments helped me a lot in not only exploring the possibilities, well it's no surprise it was 'Google Ai Studio' but this is not about which tool is "technically" better in a comparison chart; it was about which tool felt most comfortable for me to build a solution efficiently. (And no, this is not a paid post! I just genuinely enjoyed the workflow).

What's next ??

This journey from a simple prompt to a full-fledged tool has been a massive learning curve. In Part 2, I’ll share exactly how I used Google AI Studio (and a few other helpers) to actually build and launch BeamUp.in.

Happy New Year!

As we step into this new year, I hope this tools will be helpful to people in need and my story inspires you to start tinkering with your own ideas. Here’s to a year of learning, building, and breaking things until they work.

See you in the next post!