The Failures That Built Fincil

It started, as most chaotic engineering stories do, at a Hackathon.

The problem statement I chose was simple: Build a Fintech Twin. The goal was to create an AI that understands your finances as well as you do. I sat down, opened my IDE, and started coding a standard RAG (Retrieval Augmented Generation) chatbot.

But about six hours in, I realized I was failing. I wasn't building a "Twin." I was building a glorified search bar. If I asked it, "Can I buy this?" it would just look at my bank balance and say "Yes."

That’s when I realized the problem with modern fintech. It’s too passive. It treats finance like a math problem (Income - Expense = Balance). But finance isn't math. It's psychology. It's impulse. It is "Doom Spending."

We’ve observed that Gen Z, Millennials, and almost everyone with a limited income regrets a purchase at least once. We wanted to build something that removes that feeling of regret.

Think about "Digital Wellbeing." When phone manufacturers first introduced app blockers, everyone thought, "Why would I block an app I installed myself?" But now, millions use those tools to stop doom-scrolling. I believe finance needs the same mechanism. This belief is what kept me fueled, even when people told me the idea was futile.

The Pivot: From One Agent to Three (v1)

I was stuck. I was deviating from the problem statement, and the deadline was ticking. Then, my friend gave me a research sheet about the Council of Agents.

The core concept hit me immediately: One AI agent is an enabler. Three AI agents are a debate.

If you ask a single AI, "Should I buy this Phone?" it might hallucinate a justification. But what if I split the AI's personality?

• What if one agent was a Miser (obsessed with safety)?

• What if one was a Visionary (obsessed with ROI)?

• And what if I forced them to fight?

That was the birth of Fincil v1. I stopped building a chatbot and started building a Tribunal.

• The Miser (The Bear): Risk-averse. Uses vector memory to pull up every time you wasted money in the past.

• The Visionary (The Bull): Growth-focused. Checks ROI and long-term goals.

• The Twin (The Judge): Synthesizes the argument and gives a verdict.

The First Failure: "Everyone" We submitted v1 to the judges. They were amazed by the tech, but we made a crucial mistake. They asked one question: "Who is the target audience?" I panicked. I said, "Everyone." That was the wrong answer. Even as I said it, I knew I had messed up, but I couldn't take it back. We didn't win that hackathon. We stood in the Top 10, but we walked away with a lesson worth more than the prize money: If you build for everyone, you build for no one.

The Missing Link: The Human Factor (v2)

After the hackathon, while polishing v1, I realized another flaw: The Council was a dictatorship. If the Miser said "No," the conversation ended. Why should the AI have the final say? What if I have a valid reason?

This led to Fincil v2. We introduced the Rebuttal Loop. Now, if the Council blocks a purchase, you can argue back. You can say, "I cancelled my Netflix subscription to afford this." The agents ingest this new context and re-evaluate. It turns financial discipline into a negotiation.

I wanted to implement this immediately, but exams got in the way. Then, I saw a hackathon at GITAM University. I registered, hoping this would be my redemption arc.

The Second Failure: "Do You Use It?" I built the feature. I was ready. But during the event, I showed it to my friend and a friend of his. They didn't care about the complex architecture or the Rebuttal Loop. They just looked at me and asked: "Do you use your own app for financial decisions?"

It was a simple question. And I made my second big mistake. I answered honestly: "No." I tried to explain that I know my financial condition well enough without it, but my friend cut through the noise: "That's it. Even you don't use the app you built."

A sudden wave of remorse hit me. Why did I even choose to rebuild this? For a moment, I gave up. I sat there, defeated, while a guy next to me was building a simple "Student Resource Platform"—essentially a website tying links together.

I thought, "Is that even a hackathon idea?" I had previously tried tying multiple things together and was told to focus on one thing. Yet, to my shock, he won the hackathon.

I asked myself: What did I miss? Is there something I overlooked? Why do I fail at the last spot every time? I have learned a lot of things, but I still feel there is some tiny secret hiding in those winning projects that I just can't see.

The Noise vs. The Code But coming back to Fincil, something shifted. I decided I didn't want to listen nor care what anyone was telling me. I just wanted to finish the thing.

I put my head down. I added analytics, a tracker, and history features—the "boring" stuff present in all other apps just to make it complete. And finally, I finished the project.

The 3rd Round: The Stare and The Stutter Then came the 3rd round. They asked to note down the tech stack. I stood up and stated my complex stack: Llama 3.3, Groq, LangGraph, Supabase Vector Store. I don't know why, but the room went silent. I felt everyone staring at me. They were amazed. For a brief moment, I felt like I had this in the bag. The engineering was undeniable.

But then came the presentation. It was a virtual meeting, but oddly, I didn't find a panel of judges just one person watching over everything. And that is when the enemy I have been fighting since my first hackathon returned: my voice.

Even though I felt confident when I was coding, the moment I had to present, my confidence was wiped clean. My voice shuttered. All the preparation, everything I wanted to say, just evaporated.

I think this is where I lost again. I realized it immediately after the presentation. I had the code, but I didn't have the voice. I have to find a way to overcome this.

The Redemption

(The Pivot to V3) But before this, I met a person at a buildathon I attended (I will write a special blog about this soon).

While talking, he told me he actually liked Fincil. He didn't ask if I used it; he said he wanted to use it. At that moment, I felt very happy. He shared his specific financial problems with me, and that conversation struck me with the idea for the next version of the project.

I have drafted it, and I am hoping to build and release it soon.

The Takeaway

If there is one thing I learned from losing those hackathons and facing those awkward questions, it is this:

"Complexity impresses engineers, but simplicity wins hackathons. And you don't need a panel of judges to validate you; you just need one user who says, 'I need this.'"

I am still finding my voice both in the presentation room and in the product but I am not stopping until I find it.

Try Fincil v2

Today, as I promised, I will share the deployment link of Fincil v2. Hope this will be useful at some point.

🚀 Link: https://fincil.vercel.app/

In the next blog, I will provide the complete technical documentation of how we built v2.

Fincil v3 is already in the works it will be more robust, with proactive features that solve the problems v2 couldn't.

See you in the next log.

—Aneesh

---