Beyond the Prompt: Meet jaypatel1325, AI Exponential League Champion
Season 1 of the AI Exponential League ran from April to July 2026.
Many challenges. Agent workflows, evaluation harnesses, go-to-market systems, video pipelines. Different industries, different problem shapes, different ways to fail.
At the end of it, one name sat at the top of the leaderboard: jaypatel1325.
Jay is a software developer from Gujarat, India. He found Topcoder through friends and stayed because the problems kept changing. The challenge that stretched him furthest was an AI-based go-to-market build, which pushed him into deep agents and evaluation harnesses he hadn't worked with before.
We asked him how he opens a new challenge, when he walks away from an approach that isn't working, and where he thinks human judgment still matters. Here's the conversation.
Tell us a bit about yourself - where you're based, what you do, and how you first found your way to Topcoder.
Hi, I'm Jay from Gujarat, India. I'm a software developer, and I first heard about Topcoder through my friends.
What made you decide to commit to the AI Exponential League this season?
Two things. First, the challenges covered many different AI topics and were genuinely challenging . Second, I realized AI-assisted development had changed how much a single engineer could build in a short time. How cool is this :)
You finished the season at the top of the leaderboard - congratulations! Looking back, what do you think actually made the difference between you and the rest of the members?
I think the biggest difference was learning from feedback. After every challenge, I carefully read the reviewer comments and the AI review, noted my mistakes, and made sure not to repeat them in the next challenge.
The AI Exponential League spanned a lot of different AI problems across industries. Which challenge stretched you the most, and what did you have to learn on the fly?
The Build an AI-Native Go-to-Market Platform challenge was the toughest one. I learned a lot, especially about deep agents, evaluation harnesses, and agent workflows.
When you open a new challenge, what are the first things you do before you write any code?
First, I read the requirements carefully and write down all my questions. Then I make a rough plan and improve it until it is simple and easy to understand. After that, I choose the tech stack and work through the plan step by step. I also use Google AI tools to brainstorm and focus to identify the unknown parts before I start coding.
A lot of members tell us the gap between an average submission and a winning one is in the last 10% - documentation, edge cases, polish. Does that match your experience, or is the real difference somewhere else?
I agree, but I think the real difference starts much earlier. If you understand the requirements well and have a good checklist from day one, the final polishing becomes much easier.
Was there a point this season where a challenge didn't go the way you hoped? What did you take from it?
I didn't have a challenge that went badly, but I learned an important lesson: never trust an AI component without testing it. Now I validate every important AI step before building on top of it, and I keep the rest of the system simple and predictable.
How much of your edge comes from the AI itself versus how you frame the problem, structure the solution, and validate it?
For me, planning is the biggest advantage. I spend time understanding how the solution should work, brainstorm with AI, finalize the plan, and only then start building.
When you start an AI challenge, how do you decide which approach to try first - and how quickly do you abandon one that isn't working?
I always start with the simplest approach that meets the requirements, then I improve it step by step. If an approach is not working, I don't spend too much time on it. Even when one fails, I learn from it, and sometimes combining ideas from different approaches leads to a better solution.
Where do you think human judgment still matters most in building AI solutions, even as the models get stronger?
Human judgment is still very important. AI can generate ideas and code, but people still need to make the right decisions, review the results, and make sure the solution is correct.
How has the way you build AI solutions changed over the last year as the models have improved?
A year ago, I mostly used AI to help me write code faster. Now I use it throughout the development process for planning, building, and improving solutions, while I focus on reviewing and making the final decisions.
What AI skill are you deliberately trying to get better at right now?
I'm focused on deep agents and evaluation harnesses. Building AI agents is getting easier, but making sure they work correctly is the skill I most want to improve.
----------------------------------------------------------------------------------------------
Jay's approach isn't a secret weapon. Read the requirements before you write code. Write down every question you can't answer yet. Read the reviews after you submit, even when you win. Never build on top of an AI step you haven't tested.
None of that depends on which model you have access to.
Keep pushing!
Season 2 of the AI Exponential League starts in August - stay tuned for more details!