How AI Took My Journey from a Coding Beginner to Building Real Products

I’m not a software engineer. I was never supposed to be one.

The last time I wrote code was in 2004 when I learned C and C++ for my mechanical engineering degree. Fast forward to 2024, and I’ve built a directory of 790+ AI tools, including a Chrome extension, a vernacular resume builder that handles payments and a desktop app for video generation. None of this would have seemed possible for someone like me.

But here’s the thing: AI didn’t turn me into a coder. It turned me into a builder who can code when needed. And that difference matters more than you might think.

Where I Started: A Mechanical Engineer with Zero Coding Experience

Let me set the scene properly…

I’m a mechanical engineer by training, and my last experience with programming consisted of writing basic C/C++ scripts two decades ago. I had an entrepreneurial mindset, sure, but I wasn’t the person you’d expect to ship software products.

I could barely set up a development environment. The idea of deploying something to production felt like a different language. When people talked about APIs, backends or frontend frameworks, I nodded along but didn’t really understand what they meant in practice.

I had ideas—lots of them. But the gap between “I want to build this” and “I can build this” felt impossible to cross. Hiring developers was expensive. No-code tools were limiting. And learning to code “properly” would take years I didn’t have.

The Catalyst: Antler and Seeing AI’s Potential

The turning point came during the Antler residency program in Singapore during the last quarter of 2023 (Oct 2023 to Dec 2023).

Antler brings together 80+ smart people from across the globe. Founders, builders and creative minds, all in one place for an intensive program, trying to solve real problems and launch startups. The environment is intense. Everyone’s thinking at scale, pushing ideas forward.

That’s where I got exposed to what AI could really do.

People around me were trying to build products with AI, and they were doing it. For the first time, I was hearing technical concepts like “GPU” thrown into conversations naturally. But it wasn’t just the jargon—it was the ideas. The sheer creativity in how people were thinking about leveraging artificial intelligence. These weren’t people writing emails or summarizing documents. They were thinking about building entire applications with AI.

I remember the realization: “This is going to be a big thing. We can build a lot of stuff with AI.”

It wasn’t some grand epiphany. It was practical observation. Smart people around me were convinced about AI’s potential, and they were actually doing something about it. If they believed this was the path forward, maybe it was worth exploring seriously.

I returned to India after Antler with conviction. In January 2024, I joined 100xEngineers—a structured GenAI program designed to teach people how to build with AI. No theory, no hype. Just practical skills for shipping products.

I went in uncertain. But I was committed to finding out what I could accomplish.

The 100-Day Roller Coaster: From Noob to Builder

Here’s where things got real.

Midway through the program, Siddhant (the co-founder and trainer of 100xEngineers) suggested a challenge: post on Twitter every single day for 100 days about what we learned. My first thought? “That sounds like a lot of work…”

But I did it anyway. I created a small calendar, committed to the challenge and asked my wife to check in daily to make sure I stayed on track. I needed that accountability. Without it, I knew I’d quit eventually.

Excitement (Days 1-20)

The beginning was thrilling. I was learning how image diffusion models worked, experimenting with prompts and creating things I never thought I could. The material was fascinating, and my wife made sure I stayed consistent with my commitment to sharing my learning on my social platform.

I started building small projects, including a cartoon storybook tool and LoRA training for images of myself and kids. These weren’t production apps, but they were encouraging proof I could make things work.

Frustration (Days 21-50)

Then the learning got harder.

Setting up AWS servers. Debugging errors that made no sense. Spending hours on a single issue only to realize I’d misconfigured something basic. There were days when I accomplished almost nothing, and I still had to post on Twitter. What was I supposed to say? “Spent 5 hours debugging and got nowhere?”

That is when the cohort became pivotal. Fellow cohort friends were posting their progress, sharing their wins and struggles. Seeing them push forward kept me going. If they could figure it out, so could I. Sometimes, peer pressure can be positive.

Questioning Everything (Days 51-70)

Around this point, self-doubt started to creep in.

Python was new to me. The AI concepts were dense. I didn’t have a CS background. I wondered if I was fooling myself into thinking I could do this.

That’s when generative AI became critical—not just as a tool, but as a learning partner. I’d ask Claude or ChatGPT to explain concepts I didn’t understand. I’d paste error messages and get debugging suggestions. I’d ask for code and then study it to figure out how it worked.

AI wasn’t writing perfect code for me. But it was helping me understand the steps I needed to learn next.

Breakthrough (Days 71-90)

I experienced another shift around day 70.

I stopped trying to understand everything perfectly and started focusing on getting things to work. Debugging became less frustrating because I understood the patterns. I could read error messages and know where to look for solutions. I started thinking in terms of workflows—what needed to happen, in what order and with what tools.

Then came a moment of validation. One of my projects got featured on Tanmay Bhat’s show. It was an AI model of Vodafone’s ZooZoo character. The virality was unexpected. People reached out asking for advice, assuming I was an expert. I wasn’t—but I definitely knew more than I did 70 days ago.

Today, I’ve actually open-sourced the project, so if you want to test it out, check out the link here.

Momentum (Days 91-100)

By day 90, posting daily had become automatic. My wife stopped asking if I’d done my work—she’d known I had. The consistency had turned into a habit.

Finally, I completed 100 days. Did I become an expert coder or GenAI engineer? No. But I was skilful enough to build things. I knew how to leverage AI to write code, create agents and ship projects. That was enough to start building real products.

The Compounding Effect: From 100 to 365 Days of Consistency

Here’s what I didn’t expect: the 100-day discipline became a system.

The practice of showing up daily, documenting what I learned and sharing it publicly didn’t end when the challenge finished. I applied the same approach to writing on TheToolNerd and I’ve been publishing consistently for over a year now.

The pattern is clear: social commitment → accountability → momentum → consistency → results.

The 100-day challenge wasn’t just about learning GenAI. It was about building a sustainable practice of showing up, even when I didn’t feel like it. That habit has compounded—first into product development, then into consistent content creation and now into a system I use for everything I build.

Public commitment works. Not because it’s magical, but because it removes the decision-making. You’ve already committed. Now you just show up.

The New Reality: What Being a “Builder” Actually Means

So, what does it mean to be a builder who codes with AI versus a traditional software engineer?

Let me break down what I can do now—and where I still struggle.

Skills I’ve Developed

1. Problem Decomposition: I can look at a product idea and break it into components—What’s the data model? What’s the user flow? What APIs do I need? What’s the frontend vs. backend logic?
2. Prompt Engineering for Code: I’ve learned how to direct AI effectively. Instead of saying, “Build me a feature,” I say, “Write a Python function that takes a URL, scrapes the pricing table using BeautifulSoup and returns a JSON object with these specific fields.”
3. A Debugging Mindset: Most of coding isn’t writing—it’s fixing. I’ve learned how to read error logs, trace issues and iterate until something works. This took the longest to develop.
4. Architecture Thinking: I understand how pieces fit together. Database → Backend API → Frontend UI. Authentication flows. Payment gateways. Webhooks. Edge functions. I don’t know every detail, but I know the structure.
5. Shipping Bias: I’ve learned that “good enough to ship” beats “perfect in my head.” Most of my products launched at 70% and improved from there.

Real Solutions I’ve Built

Let me show you what this looks like in practice.

ToolNerd Library

This was my first real product, built in October-November 2024. It’s a directory of 790+ AI tools with automated scraping.

Here’s how it works: You enter a URL, and the app fetches screenshots, pricing, features and relevant YouTube videos. The backend is Python FastAPI, the frontend is Next.js and the database is SupaBase. I integrated OpenAI for scraping and analysis, YouTube APIs for training videos and built an automation pipeline using Make.com and Slack.

I didn’t know FastAPI when I started. I learned it by building this product. The first version took weeks because I had to figure out how to structure the backend, handle API calls, format JSON responses and display them properly on the frontend.

The app generates some revenue—not from affiliate links (which haven’t worked well)—but from writing reviews on G2 and similar platforms, which covers hosting costs.

I also built a Chrome extension version. The idea was to get people to see new tools every time they opened a tab, rather than visiting the website. Distribution has been a challenge—but more on that later.

HelloAryan

This is a vernacular resume builder. People speak in Telugu, Hindi or Tamil, and the app generates a professional resume in English.

I started building this during a Bolt hackathon. The UI wasn’t working, so I moved to Lovable, got a better design and then pulled everything into Claude Code for final development. The stack: React frontend, OpenAI for language processing, Python backend and SupaBase edge functions for AI logic.

The hardest part was PDF generation. I needed the resume to match the preview exactly. I ended up using PlayWright to render the React view into HTML, then convert that into a PDF. It took days to get right.

The second challenge was dynamic UI. On the left side of the screen, users chat with the AI. On the right side, the resume updates in real time. Making the layout adapt to different amounts of content—especially for experienced professionals with 5-6 page resumes—was harder than I expected.

This was my first end-to-end SaaS product with payment gateway integration. I learned how to handle authentication, user accounts, subscriptions and payment flows. It’s live and functional, though distribution remains an issue.

With one workflow I built, I can upload a UI design image, and a Claude Code slash command generates the React template and HTML file for PDF rendering. This saves me hours when adding new resume templates

ConvertHub

This project started as a personal need. My dad is a trainer and consultant who creates lots of presentation content. He recorded videos for Udemy but wasn’t comfortable on camera. I thought, “What if I could convert his presentations into videos using his voice?”

That’s how ConvertHub was born. The main feature, SlideCast, converts PDFs to videos. There’s also an AutoDub feature where you can upload a video, and it dubs it in your voice by analyzing the video, writing a script and generating a voiceover.

I tried building this as a web app first, but PDF-to-video rendering didn’t work because of FFMPEG requirements and CPU capacity limitations. I realized rendering is resource-intensive, and desktop apps exist for a reason. So, I rebuilt it as an Electron desktop app with a lifetime license model.

The app supports bring-your-own-key integration—Gemini, OpenAI, Cartesia and ElevenLabs for voice cloning. I built it mostly with Factory.AI, which I found underrated and effective for solving complex problems quickly.

The website landing page was built with v0 and is hosted on Vercel. It was my first time deploying on Vercel, and I finally understood why people say it’s beginner-friendly. To put it plainly, it just works.

Other Tools

• ToolNerd Chrome Extension: “New Tab” displays a random AI tool for discovery
• Itsmyclub.vercel.app: Build your own persona and interact with it
• Designstyles.xyz: A reference guide for design style names

What I Haven’t Mastered Yet

Let me be honest about the limits.

I’m not a software engineer in the traditional sense. If you asked me to optimize a database query, architect a microservices system or debug a concurrency issue, I’d struggle. I don’t know most computer science fundamentals. I can’t write efficient algorithms from scratch.

But here’s the key: I don’t need to know everything to build products that work. AI fills the gaps—not by doing everything for me, but by helping me learn what I need, when I need it.

Lessons for the Skeptics

I know what you’re thinking. “This sounds too easy.” “What about code quality?” “What about hitting a wall without fundamentals?” “Isn’t this just copying code?”

Let me address these directly.

"Won’t AI-generated code be terrible?"

Sometimes, yes. And you’ll learn to fix it. That’s the skill—not writing perfect code, but debugging imperfect code until it works.

"Won’t you hit a wall without CS fundamentals?"

Probably. But by the time I hit that wall, I’ll have shipped 10 things. And I’ll know exactly what I need to learn next because I’ll have encountered the problem in practice.

"Isn’t this just copying code without understanding?"

No. It’s collaborative problem-solving. I’m directing AI, reviewing what it generates, modifying it when it’s wrong and learning from the process. That’s not copying—that’s using a tool effectively.

The Real Lessons I Learned

1. Momentum beats perfection. I shipped products at 70% completion and improved them based on real feedback. Waiting for perfection would have meant shipping nothing.

2. Debugging is the real skill. Writing code is 20% of the work. Fixing code is 80%. If you can debug, you can build.

3. Connecting the dots happens over time. Many things didn’t make sense initially. By day 100, they all connected. You need patience for this part.

4. Social commitment accelerates learning. The 100-day Twitter challenge forced me to show up daily. Without that public accountability, I would have quit when things got hard.

5. Distribution is harder than building. I can build products now. Getting people to use them? That’s the part I’m still figuring out. Technical skill doesn’t solve distribution problems.

6. Tools evolve constantly. AI models improve. New tools launch. MCP servers, Claude Code slash commands, Cursor features—staying updated matters. The learning doesn’t stop.

The Path Forward

I’m not done learning. This is just the beginning.

The AI landscape changes every month. New models launch. New capabilities emerge. What works today might be outdated in six months. The only way to stay relevant is to keep building, keep experimenting and keep learning.

My next challenges aren’t technical—they’re distribution. I can build products, but can I get them in front of the right people? Can I turn users into customers? Can I build sustainable businesses around these solutions?

Those are different skills. I’m learning them the same way I learned to code: by doing, by failing, by iterating.

If a mechanical engineer who last coded in 2004 can build SaaS products, what’s stopping you? The tools are here. The models are capable. The barrier isn’t talent or background—it’s whether you’re willing to commit 100 days to figuring it out.

You don’t need a CS degree. You don’t need years of experience. You need curiosity, persistence and the willingness to debug when things break.

Start small. Build something. Ship it. Learn from it. Repeat.

That’s the process. It’s not glamorous, but it works.