This is a submission for the Bright Data AI Web Access Hackathon
🔥 What I Built
OpinionFlow helps users skip the endless scroll. It'...
Some comments have been hidden by the post's author - find out more
For further actions, you may consider blocking this person and/or reporting abuse
Super cool seeing live reviews merged with quick AI insights like this. Any plans to add more stores or deeper personalization next?
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.
Top-notch work! Sir
Thanks for sharing this sir!
Very great work❣️
Great work as always!!
Thanks for always showing up and supporting. Means a lot!
Fabulous this is! Such a life saver. Good work!
Thank you so much! Glad to hear it's helpful to you. Let me know if you try it out!
Good work
Thank you
Honestly didn’t think something like this could be built in a hackathon timeframe. Hats off!
Great work shivansh, thanks for sharing this
Really appreciate that - it was a fun challenge putting all together.
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?
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.
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!
I'm glad you liked it! Thanks for commenting.
What stood out to me is how you have aligned semantic caching with product discovery - it felt like the kind of detail that only comes from actually building and testing products deeply.
Thank you again!
Woow, appreciate that insight - yes caching was the trickiest part in this. Took a few iterations to get the key structure right. Pinecone is great too
Would love to see multilingual support in future, you are currently looking for amazon.com but they have different variants for different countries. So maybe you can try to look at or detect user's location and then connect to amazon or walmart. BTW! really great work.
Was it tough to parse the HTML across different stores like Walmart and Amazon?
100%. Every store has different DOMs, lazy loading, and anti-bot tricks. The use of browser api, and other servies from Brightdata made it easy to extract information.
Sometimes the website shows, error loading analysis
Happened with me too! But then I guess now it's working.
This is incredibly helpful for comparison shoppers like me! Any plans to add Flipkart or BestBuy next?
The scraping structure is modular, so Flipkart and BestBuy are 90% ready - just need to fine-tune HTML selectors and test a few edge cases.
Tried searching for 'MacBook Air' — worked beautifully. Would be great to export reviews too!
Great to hear that! Expoert is a solid idea - CSV or PDF maybe? I'll see how I can fit it into the next sprint.
Would love to see how it handles products with thousands of variants. Great job Shivansh!
That's definitely a trick part! Right now, it clusters reviews at the product level, but doesn't differentiate by variant (like size or color).
Maybe in future versions, you could allow users to contribute their own reviews and see how Gemini classifies them?”
Ooh I really like that! Like a sandbox mode for testing your own feedback - thanks for the idea, bookmarking it.
Brooo this is insane 🔥🔥 I remember when you were just sketching this out — now it’s a full-blown AI system with caching and scraping and everything. Let’s gooo!
Man that means a ton! You've seen the whole journey, from doodles to delivery. Appreciate you cheering me on the way!
Just used the live demo — the way it breaks down pros/cons and sentiment across stores is honestly more helpful than most YouTube review videos lol.
Hahah I'll take that as a big win. Glad you liked it!
One suggestion: showing the last scraped timestamp for each store would build a lot of user trust. Just a little ‘Fresh as of…’ badge maybe?
Love this and totally agree. Adding a "Last updated: X mins ago" badge per store is actually super simple to implement with current cache structure. Will try to ship that next !
That semantic caching with Hugging Face is super interesting. Do you embed each review separately or batch them by product?
Great question - I batch them by product for now, since it makes Gemini's job easier too. Thinking of switching to per-review embedding for more granular analysis later on.
Loved the demo! Would be cool to know how Pinecone handles paraphrased or fuzzy queries. Does it match on semantic similarity only?
Yes exactly - it works on embeddings, so even if oyu say "best wireless earbuds" and someone says "top bluetooth headphones", it knows to match based on meaning, not words. Super powerful!
Quick question: does the Gemini analysis handle multilingual reviews? Or is it limited to English for now?
Currently it's mostly tuned for English, though Gemini does handle multilingual reasonably well. Planning to test with Hindi and Spanish next to see how it holds up.
One idea: could be awesome to let users upvote or bookmark the most relevant insights per product. Adds a human layer on top of AI
Yes! That would make it more community driven and help get most out of it.
Absolutely loved the clarity of the summaries — way better than skimming 100+ reviews. You’ve nailed both utility and presentation
Thanks so much - took a tweaking to make Gemini return clean and accurate summaries. Glad it's working as intended!
This is the kind of thing I’d actually use before buying anything online. Let me know when you launch a full version!
That made my day! I;m polishing the UI and will definitely drop an update once it's product-ready.
Хөөе энэ үнэхээр гайхалтай.
Баярлалаа! Thank you so much — appreciate the love!
Great work
Thank you! Really appreciate your time to check this out
One thing I’d love to see is a timeline feature — like how sentiment shifts over time for a product as updates roll out or versions change.
Yes yes yes — time-based review analysis is one of my favorite ideas. Currently the backend already stores timestamps per review, so i can do sentiment shifts over weeks/months. Thanks for commenting
Would it be possible to let users upload a product URL from any store and get analysis instantly? That would make this even more flexible.
Yes! I've already built in support for direct URLs. Would be live soon!
I didn’t even know Pinecone could be used like this. Thanks for showcasing how semantic search can power something practical like this.
Absolutely! That's what I was aiming for - real-world usage that just goes beyond the demos. Happy you found it interesting!
Honestly surprised at how polished this is. The backend flow makes sense and the front-end is clean too. Respect.
That means a lot - thank you! It took a lot of iterations to get it this clean. Happy the backend/frontend flow felt smooth to you.
Great work. I also applied to this hackathon, but couldn't finalize the end product. Great to see how you finalized this project.
Totally get that - sometimes it just doesn't click in time. Hope you jump back into the next one!
How you integrated Amazon review scrapping in this? BTW project looks solid.
You can find Amazon Review extractors in Bright Data - Web Scraper for Amazon
Оцените пошаговое руководство. Пользовательский интерфейс интуитивно понятен и не кажется перегруженным.
Большое спасибо! Appreciate the feedback — I worked hard to keep the UI clean and beginner-friendly.
Curious how LangChain fits into the final RAG flow — are you using custom retrievers?
Yes- I'm using Langchain to orchestrate the retrieval and prompting. The retriever pulls from Pinecone using cosine similarity and then fiilters based on metadata like store+timestamp before the Gemini call.
Just saw your linkedin post, and wanted to really thank you for this creation.
Question: How do you manage inconsistencies in review formats between stores? I imagine Amazon and Walmart have very different structures.
This feels like ChatGPT with a memory — but for ecommerce reviews. Great job!
Could be intereting to integrate price tracking next. This way intelligence + price tracking would be really good.
The way it summarizes real user sentiment into a one-liner is impressive. Curious — how long does it take from query to final answer?
Great project, wishing you luck for this hackathon!
Super clean interface - loved it