Overview
Designed an RAG-based Mutual Fund FAQ Assistant for the Groww ecosystem that delivers facts-only, source-backed answers to mutual fund queries. The assistant retrieves information exclusively from trusted public sources, ensuring every response is accurate, verifiable, and compliant. The focus was on building a trustworthy AI experience that simplifies information discovery without providing financial advice or recommendations.
AI Tools
Stitch by Google
Cursor
Objective
Design and build a lightweight Retrieval-Augmented Generation (RAG) assistant that:
-
Answers factual mutual fund questions
-
Retrieves information only from official public sources
-
Provides concise, source-backed responses
-
Avoids investment advice, opinions, or recommendations
Target Users
-
Retail investors researching and comparing mutual fund schemes
-
Customer support teams handling repetitive mutual fund queries
-
Content and operations teams requiring quick access to verified information
Summary
Create a reliable AI assistant that enables users to quickly find accurate, verifiable, and compliant information about mutual fund schemes. By grounding every response in official sources and eliminating subjective recommendations, the solution improves trust, reduces misinformation, and enhances the overall information retrieval experience.
Understanding the concept
Before jumping into wireframes and UI, I wanted to understand how a Retrieval-Augmented Generation (RAG) system actually works. I learned that a RAG system doesn't rely only on the language model's knowledge. Instead, it retrieves relevant information from a trusted knowledge base before generating a response. This makes answers more accurate, transparent, and grounded in official sources rather than assumptions.
The quality of the assistant depends on the quality of its data. I explored how documents should be cleaned by removing unnecessary content, standardizing formats, and retaining only reliable, relevant information. Clean data leads to better retrieval and fewer incorrect responses. I learned that good chunking helps the retrieval system find the most relevant information quickly, while poor chunking can result in incomplete or confusing answers.
This project shifted my perspective from designing only user interfaces to designing the entire AI information retrieval experience. Understanding how data is prepared, retrieved, and transformed into trustworthy responses helped me make more informed product and UX decisions.
Building with Cursor
I built the application using Cursor, leveraging AI to accelerate development, debug issues, and iterate on the product. I first defined the information architecture to organize user queries and retrieval flow, then cleaned and structured data from official sources for accuracy. Using AI prompts, I optimized document chunking to improve retrieval quality and chatbot responses. Finally, I tested and deployed the end-to-end RAG application, ensuring users receive fast, accurate, and source-backed answers.
Final UI — Refined Experience
I designed the user interface in Stitch, focusing on a clean and intuitive chatbot experience. I refined the layouts, interactions, and response patterns to ensure clarity and ease of use. Once the interface was integrated and tested, I deployed the application on Vercel, making the end-to-end prototype accessible and ready for demonstration.
→ View HDFC mutual fund chatbot
