Describe your project

Here’s a breakdown: 1. What’s In-Scope: Conversational AI at its core: Our agents are designed to handle common BFSI queries—whether it’s about comparing insurance policies, managing claims, or even assisting with banking services. Think of Priya, our AI assistant, available 24/7, guiding users through complex stuff like policy selection or answering banking inquiries in real time, and all with a tone that’s helpful, not robotic. Human-like Conversations: This is key. Our agents aren't just throwing information at users—they’re actually conversing, responding based on the user’s unique situation and questions. We’ve engineered them to engage in meaningful conversations, which builds trust and fosters better customer relationships. Automating the boring stuff: Tasks like form-filling, repetitive claims inquiries, or even onboarding customers can be handled seamlessly by our AI agents. It’s all about making these processes faster, reducing errors, and giving the user a smooth, easy experience. Tailored for BFSI: We’re laser-focused on insurance, banking, and financial services. That means our solution understands the specific pain points in this sector—like regulatory stuff, detailed policy questions, and financial product recommendations. While we don't focus on advanced fraud detection or compliance-heavy analytics, which require specialized security tools, our primary focus is on enhancing customer experience. We also don’t replace human financial advisors for complex financial planning, but our AI assists with policy selection and account management. Looking ahead, we see huge opportunities in multilingual support for India’s diverse market, predictive assistance that anticipates customer needs, and cross-industry potential, with applications in healthcare and e-commerce, making this tech adaptable beyond BFSI.

Challenges I ran into

As a product designer (not a coder), one of the toughest challenges I faced while building Priya, was getting the AI to handle complex, industry-specific conversations in a natural, human-like way. I started with pre-built AI models, but they struggled to deliver the personalized, deep responses needed for various insurance products. The conversations felt too generic and robotic, and didn’t match the kind of experience I wanted to create for our users. I leaned heavily on low-code and open-source tools to get the job done. I used a mix of platforms that let me customize the AI’s conversation flow. One key thing I did was train the AI using real-world data from Policy Bazaar’s interactions. This helped make the responses feel much more tailored, improving its ability to handle specific questions about things like health or motor insurance. Another big hurdle was setting up a Retrieval-Augmented Generation (RAG) system to dynamically pull insurance policy information from various pages on Policy Bazaar. The traditional web scrapers I tried initially kept breaking whenever a new page or sub-link was added, making them unreliable for the dynamic updates I needed. So, I hacked the process using Make, an automation tool that allowed me to set up webhooks to scrape data from multiple Policy Bazaar sub-links. The webhooks collected data and pulled it into a central knowledge hub that powered Priya’s ability to respond with accurate, up-to-date information across different insurance categories. Getting those webhooks to pull clean, reliable data wasn’t easy—it took a lot of trial and error. But once it worked, it meant that Priya could reference the latest info without needing manual updates whenever a new insurance policy or page was added. In the end, I was able to build a scalable, dynamic AI assistant using low-code tools, some creative problem-solving, and real-world data—showing that you don’t need to be a hardcore coder to create powerful solutions!