EchoReview

Website

Inspiration

It all started when my teammate told me about the time he was getting a gaming PC, and he was tasked with the decision of choosing the most appropriate one. Like any normal human, he scoured the internet reading numerous websites and watching videos to get reviews of specific models. I realized how many people are extremely particular about what they buy and how much they take into consideration before purchasing, giving me the idea to build something upon that and make it easier for everyone.

What it does

When a URL of a product page is entered, it scrapes the product page and identifies the product in question, and a custom search engine is employed to scrape 40 relevant websites from the web with different reviews and perspectives, using a pre-trained LLM model to compile the results and present them on a website simply and elegantly.

How we built it

This system works on an interlinked Node.js and express backend server and the Vite and Vue.js frontend. The Scraper API is used to gain the HTML data of the product page which is parsed and analyzed to find the product name. The Programmable Custom Search Engine of Google was used to search for the top websites that contained reviews from users and companies to create a collection of reviews. This collection was fed into a pre-trained llama model by Meta, which was used to output pros, and cons, a detailed review, and a rating to display to the users. The Vue front end was styled using Tailwind CSS to make it simple yet appealing. The JS API of the Vite platform was used to run the development server and build the production-ready environment, using its superior server-side rendering to enhance the speed of the website, despite the volume of scraped data. This backend is also connected to a separate Chrome Extension that detects product pages by default and is activated to immediately display the collated reviews about the product.

Challenges we ran into

Our journey was replete with challenges throughout. The first challenge we faced was parsing the data we received from the web scraping and getting the appropriate HTML tags from the data. Following that, it was hard to accurately scrape the appropriate websites and use them to gain insights in a manner that would present original opinions and not just AI jargon. A common challenge we faced was rate-limiting while also facing repeated issues with production-ready environments and deployment due to dependencies being incompatible with Vercel which made it harder to test.

Accomplishments that we're proud of and our learnings

Our team is proud of the way that we tackled all the challenges we faced throughout, as well as picking up all of the intricacies and troubles of building a full-fledged Chrome extension alongside the main website. We also managed to separate ourselves from the ongoing notion that AI projects are inherently the most complex and socially useful, by building a unique app that would help millions, while still incorporating AI where needed to speed up the process. We learned new technologies with Vite and Vue.js while finding ways to integrate them smoothly with our backend API using scrapers and even creating a custom search engine. Along with this, we enhanced our soft skills in teamwork, splitting of work, being empathetic, and learning from each other.

What's next for EchoReview

Using the TranscriptsAPI to scrape from Youtube videos as well as they are a main source of information about products as well as using ads to gain revenue to support the hosting. Furthermore, the same algorithm can be expanded to review movies, books, songs and anything that people ar interested in giving us a lot of room to build and experiment.