OmniAI

Inspiration

Our inspiration came from the common frustration with e-commerce chatbots. While many can handle basic questions, they often fail to intelligently guide a customer through a conversation, especially when it comes to product discovery and sales.

Building a truly helpful, multi-step chatbot is a complex and time-consuming task for businesses. We were inspired to solve this problem by creating a unified platform that automates not just simple replies, but entire customer journeys—from a user asking a question to finding the perfect product. Our goal was to use the power of AI to transform a rigid chatbot into a seamless, conversational sales tool.

What it does

Our product is a comprehensive, AI-powered chatbot solution for e-commerce. It automates and streamlines the entire customer journey, from initial inquiry to purchase.

It does three main things:

It intelligently handles inquiries: The system uses context-aware keyword triggers to provide instant, relevant auto-replies while avoiding unnecessary responses.
It enables seamless product discovery: Customers can find products by typing a description or by simply uploading a product image, which our system recognizes using OCR to provide a direct purchase link.
It simplifies workflow design: Our AI Conversation Designer allows users to rapidly generate and publish complex, multi-step conversation flows, transforming a static FAQ into a dynamic and interactive experience.

Ultimately, it helps businesses turn every customer conversation into a conversion.

How we built it

We built our product by first establishing a solid foundation with the AWS Kiro Spec Mode Steering Document. This document served as our single source of truth, allowing us to align all user stories and functional requirements with our technical specifications from the very beginning.

This initial alignment was crucial and enabled us to use vibe coding, a core methodology in the Kiro framework. It allowed our team to start development rapidly and with a shared understanding, eliminating early-stage friction and accelerating our progress.

For our core technology stack, the backend was built on Java 21 using the Spring Boot 3.2.5 framework, with a persistent connection for real-time communication. Our AI services integrate with AWS Bedrock for natural language understanding and OpenAI's GPT-4 Vision API for image analysis. We also prioritized security by using AWS Secrets Manager for all sensitive credentials. This meticulous approach allowed us to meet our strict performance benchmarks, ensuring chat responses are delivered in under three seconds.

Challenges we ran into

Ensuring OCR Accuracy: A significant challenge was ensuring the accuracy of our Optical Character Recognition. Since we rely on a third-party model, its performance can be unpredictable, sometimes leading to inaccurate product matches. To address this, we recognized the need to train a supplementary model, such as a hallucination detection model, to validate the OCR output and significantly increase the reliability of our product recommendations.

Maintaining Real-time Performance: With external API calls for every query, a major hurdle was meeting our strict performance requirements. To deliver chat responses in under three seconds, we focused on code optimization and efficient data handling, ensuring our system remains fast and responsive even under load.

Accomplishments that we're proud of

First and foremost, we successfully built a complete, end-to-end solution that unites multiple powerful functionalities into a single product. Rather than creating fragmented tools, we seamlessly integrated auto-replies, image-based product search, and our AI conversation designer into one cohesive system.

We are also incredibly proud of our development process. By leveraging the AWS Kiro framework, we were able to move from initial concept to a fully functional product in a very short amount of time. This showcases our team's ability to execute a complex project under pressure and deliver a high-quality result.

Finally, we're proud of the real-world value we created. Our AI Conversation Designer is a significant innovation that empowers businesses to rapidly generate and deploy complex chat flows, directly addressing a critical pain point in the market.

What we learned

Our biggest learning was the immense value of starting with a strong foundation. By leveraging the AWS Kiro Spec Mode Steering Document, we learned that a well-defined plan from the very beginning is the most critical factor in a project's success, enabling our team to move from concept to execution with incredible speed and zero friction.

We also learned that integrating with AI services is about more than just calling an API. It's about building a robust system around those services to ensure reliability, security, and consistent output, especially when dealing with the unpredictable nature of models like those used for OCR.

Finally, we learned that the true value of a product lies in its ability to create a seamless, end-to-end user experience. By focusing on integrating all of our features into a single, cohesive solution, we were able to create a product that is more powerful and valuable than the sum of its parts.

What's next for OmniAI

We've built a strong foundation, and our next step is to expand the capabilities of our AI Conversation Designer.

We plan to develop a new use case that will allow businesses to rapidly generate interactive forms for product reviews or user surveys. Imagine being able to describe the type of feedback you need, and our system will automatically create a dynamic survey form within the chat.

The vision is to go beyond simple generation. We will create a unified system to store and manage these forms, providing an automated experience similar to Google Forms, but built directly into the chatbot. This will allow businesses to effortlessly collect valuable user data, transforming the chat into a powerful tool for feedback and insights.