A.I.R.A - A.I. Retail Assistant

Leveraging voice-recognition and AI to help navigate websites and elevate the shopping experience for customers.

Comment

image

Inspiration

Have you ever found yourself overwhelmed by the deluge of information on a website? This experience drove us to create a click-less interaction with retail websites, mirroring the in-person shopping experience by engaging with an AI retail associate.

What it does

Introducing AIRA - A.I. Retail Assistant, an innovative web browser extension that elevates online shopping. It provides users with a seamless, voice-driven interface to interact with retail websites, and more.

How we built it

AIRA harnesses the power of OpenAI APIs to intelligently scan webpages and highlight relevant information based on voice queries. By leveraging OpenAI's ChatGPT and its function calling feature, we decipher user intent. AIRA can then navigate to the correct webpage or locate the desired information. Our algorithms automatically break down each webpage into sections and group together relevant data so ChatGPT understands the content. To keep operating costs at a minimum, we utilized the built-in speech recognition functionalities of web browsers. While OpenAI's Whisper was considered, we found the browser's speech recognition to be sufficiently accurate for our needs. We wanted AIRA to offer a lifelike and friendly voice, so we integrated ElevenLabs' generative AI text-to-speech service. This allows users to hear AIRA as if they were engaging in a conversation with a real person.

Challenges we ran into

  • User Intent Inference: Accurately deducing user intent, especially when navigating webpages or referring to the ChatGPT context, posed a significant challenge. We used ChatGPT function calling and context engineering to improve the AI's accuracy.

  • Codebase Refactoring: We had to refactor our codebase multiple times to create a more modular and flexible framework, making it easier to customize the user experience.

  • Asynchronous Requests and Chrome Runtime Messages: Interacting with asynchronous requests, Chrome runtime messages, and Chrome storage was complex. We initially encountered limitations with the storage of website parse trees in Chrome storage and had to find alternative solutions.

  • Optimizing Website Parse Tree: To ensure that our website parse tree remained within the maximum 16,000 tokens for GPT requests, we tried three different approaches for our parsing algorithm. Ultimately, we arrived at a solution that involved grouping together text from containers of a sufficiently low size, efficiently and effectively organizing the data for ChatGPT to understand.

  • Video Editing: Our original setup for the video editing software for the presentation was giving us graphical errors so we had to swap some responsibilities around near the end in order to get everything ready by the deadline.

Accomplishments that we're proud of

  • Customizable AI guides: We built a modular system for easily tailoring the AI to individual businesses, customizing the onboarding process AIRA guides customers through, and extracting data from users' responses to prompts.

  • Personality: We met our stretch goals and gave AIRA an animated robot mascot that's synced with ElevenLabs' TTS.

  • Working Prototype: We successfully developed a working prototype of AIRA, demonstrating the feasibility of our concept.

  • Innovative Use of Generative AI: We found a novel application for generative AI, addressing a real-world problem in a new and innovative way.

  • High-Value Concept with User Experience Focus: We created a concept that adds substantial value to users while emphasizing a user-friendly and engaging experience.

What we learned

  • Team Collaboration: We gained experience in working effectively as a team, fostering cooperation and coordination among team members.

  • Applying OpenAI Functions Dynamically: We learned how to apply OpenAI functions to dynamically react to user input, ensuring that AIRA responded effectively.

  • Utilizing Collaboration Software: We became proficient in using collaboration software such as Miro and Notion, and Git enhancing our project management and communication.

  • Project management: We practiced organizing and prioritizing tasks, distributing them among the team, and regularly meeting to check progress and adjust our plans.

What's next for A.I.R.A - A.I. Retail Assistant

  • Expanding to Other Websites: AIRA could be expanded to work with websites within the Verizon network and potentially extend its services to other businesses.

  • Reimagining as a Web Library: AIRA can be implemented as a toolkit to give businesses the power to customize how the AI guides customers through their websites.

  • Enhancing Knowledge Base: We aim to give AIRA a broader knowledge base, enabling it to provide even more comprehensive information about company products and services.

Built With

+ 3 more
Share this project:

Updates