DEV Community

Cover image for OpinionFlow โ€“ Real-Time Review Intelligence with Bright Data, Gemini 2.0 & Pinecone
Shivansh Singh
Shivansh Singh

Posted on Edited on

OpinionFlow โ€“ Real-Time Review Intelligence with Bright Data, Gemini 2.0 & Pinecone

This is a submission for the Bright Data AI Web Access Hackathon


๐Ÿ”ฅ What I Built

OpinionFlow helps users skip the endless scroll. It's an AI-powered review assistant that:

  • Crawls live product reviews from Amazon and Walmart using Bright Data MCP
  • Summarizes insights using Gemini Flash
  • Caches results semantically using Pinecone
  • Lets users ask product-specific questions via LangChain

Want to know if the AirPods Pro have battery issues? Just ask โ€” youโ€™ll get instant, evidence-backed answers.


๐Ÿ“Œ Demo


๐Ÿง  How It Works

Step Description
๐Ÿ” Search Users enter a query (e.g., "Noise Buds X Prime")
๐Ÿ“„ Crawl Bright Data scrapes live reviews from both stores
โœจ Summarize Gemini Flash extracts sentiment, pros/cons, and key specs
๐Ÿง  Cache MiniLM embeddings stored in Pinecone to avoid repetition
๐Ÿ’ฌ Answer LangChain generates natural-language responses with citations

Product discovery UI


๐Ÿงฉ Features at a Glance

Feature Description
๐Ÿ’ฌ Instant AI Answer Box Summary with links to real reviews
๐Ÿ‘ Top Pros / Cons Highlighted from verified buyers
๐Ÿ“Š Sentiment Comparison Side-by-side scores from Amazon & Walmart
๐Ÿท๏ธ Aspect Mini-Charts Dynamic breakdowns: battery, comfort, etc.
๐Ÿงญ Multi-Store Tabs Compare similar SKUs across platforms
๐Ÿ”Ž Review Explorer Drill down into sources and keywords

Sentiment analysis


๐Ÿš€ Bright Data in Action

MCP Tool Role
SERP API Fetches product listings via Google Search
Web Unlocker Unblocks product pages seamlessly
Scraping Browser Renders full review sections
Browser API + Playwright Handles dynamic navigation like "See all reviews"

Bright Data saved me dozens of hours โ€” no captchas, no proxy headaches, just clean data.


๐Ÿง  Architecture Overview

System Architecture Diagram

OpinionFlow is built with a modular microservice-style architecture with FastAPI as the core backend. Key flows include:

  • Product discovery โ†’ via Bright Data SERP API
  • Review scraping โ†’ using custom extractors
  • Analysis โ†’ powered by Gemini Flash
  • Semantic caching โ†’ via Pinecone vector search
  • Natural Q&A โ†’ powered by LangChain and Gemini

๐Ÿงฐ Tech Stack

FastAPI, React.js, Bright Data MCP, Gemini Flash, Pinecone, LangChain, HuggingFace MiniLM, Netlify, Cloud Run


๐Ÿ”ฎ What's Next

  • Add Target.com integration
  • Launch a real-time price tracker
  • Let users add their own reviews
  • Enable Gemini-powered follow-ups in chat

๐Ÿ™ Thanks

Big thanks to Bright Data and DEV for the challenge. If you liked this project, consider checking out:

๐ŸŒŸ github.com/luminati-io/brightdata-mcp

Built with โค๏ธ by @shivanshsinghh


๐Ÿท๏ธ Tags

#BrightData #Hackathon #AI #LLM #Gemini #LangChain #Ecommerce #ProductReviews #FastAPI #React #Pinecone #SemanticSearch

Top comments (78)

Collapse
 
dotallio profile image
Dotallio •

Super cool seeing live reviews merged with quick AI insights like this. Any plans to add more stores or deeper personalization next?

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Thank you so much for your comment. Yes I am thinking to add more stores and do deeper personalizations in this project. This project really resonates with me.

Collapse
 
divyanshsinghh profile image
Divyansh Singh •

Top-notch work! Sir

Collapse
 
divyanshsinghh profile image
Info Comment hidden by post author - thread only accessible via permalink
Divyansh Singh •

Thanks for sharing this sir!
Very great workโฃ๏ธ

Collapse
 
shagun_singh_ac41877ec7d8 profile image
Shagun Singh •

Great work as always!!

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Thanks for always showing up and supporting. Means a lot!

Collapse
 
scott_1209 profile image
Scott Peralta •

Fabulous this is! Such a life saver. Good work!

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Thank you so much! Glad to hear it's helpful to you. Let me know if you try it out!

Collapse
 
jennysa21 profile image
Jenny Sa •

Good work

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Thank you

Collapse
 
daliudi profile image
Dalimin Iswahyudi •

Honestly didnโ€™t think something like this could be built in a hackathon timeframe. Hats off!

Collapse
 
bahmansimonn profile image
Bahman Simon •

Great work shivansh, thanks for sharing this

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Really appreciate that - it was a fun challenge putting all together.

Collapse
 
nathan_tarbert profile image
Nathan Tarbert •

pretty cool seeing tools get tied together like this - makes me curious, you think momentum on stuff like this depends more on habits or just chasing little wins day by day?

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

Thanks! Honestly, I think it's a mix of both - I definitely chase small wins to stay motivated, but a lot of it also just comes from long-term habits. I've been coding since I was a kid, so some of that momentum is just muscle memory at this point.

Collapse
 
nitya_bhardwaj_7189adc360 profile image
Nitya Bhardwaj •

This is such a cool use of Ai and web scrapping.. something like this can save so much time and efforts of users. Great stuff man!

Collapse
 
shivanshsinghh profile image
Shivansh Singh •

I'm glad you liked it! Thanks for commenting.

Some comments may only be visible to logged-in visitors. Sign in to view all comments. Some comments have been hidden by the post's author - find out more