In a decisive move, Perplexity abandoned four months of work on their original project to focus entirely on the challenge of building a true answer engine for the web.
I love using Perplexity but knowing how the business made architectural decisions to optimize the business but also deliver the best user experience is enlightening. I feel like I learned so much about RAG and Perplexity as a business, thank you!
If Perplexity uses Amazon Bedrock to easily plug these different third-party models into their system, it is not clear to me why they need to have their own inference stack, for hosting those models not available in Amazon Bedrock?
a noob question - "Perplexity’s crawling and indexing infrastructure tracks over 200 billion unique URLs / how Vespa db is used as the index". Does it mean that they don't rely on any search api (ex: Bing api) to retrieve? I'm just amazed at this part b/c in addition to the scale problem, the data / human labeling to build this search ranking model would be huge effort
Have been using Perplexity for about a year now- really happy I have a lot on at work I have used it to build out whole community development models and programs , great to understand a bit of how it works and why it really seems more intuitive than other AI tools like co pilot etc
Your technical breakdown of Perplexity's RAG pipeline really highlights why Google is facing such intesne competition now. The fact that Perplexity can build a competitive search experience with just 38 engineers by leveraging Vespa shows how much the infrastructure playing field has leveled. What strikes me is that Google's massive web index and PageRank algorithms might not be the moat they once were if the real value shifts to retrieval quality and LLM orchestration.
i love these deep dives into how entire specific apps/sites work
I love using Perplexity but knowing how the business made architectural decisions to optimize the business but also deliver the best user experience is enlightening. I feel like I learned so much about RAG and Perplexity as a business, thank you!
If Perplexity uses Amazon Bedrock to easily plug these different third-party models into their system, it is not clear to me why they need to have their own inference stack, for hosting those models not available in Amazon Bedrock?
Thank you, this was so enlightening!
a noob question - "Perplexity’s crawling and indexing infrastructure tracks over 200 billion unique URLs / how Vespa db is used as the index". Does it mean that they don't rely on any search api (ex: Bing api) to retrieve? I'm just amazed at this part b/c in addition to the scale problem, the data / human labeling to build this search ranking model would be huge effort
Have been using Perplexity for about a year now- really happy I have a lot on at work I have used it to build out whole community development models and programs , great to understand a bit of how it works and why it really seems more intuitive than other AI tools like co pilot etc
This is a good deep dive. Not just Perplexity, coming from Amazon/Alexa background I could relate to most of the content in here.
Your technical breakdown of Perplexity's RAG pipeline really highlights why Google is facing such intesne competition now. The fact that Perplexity can build a competitive search experience with just 38 engineers by leveraging Vespa shows how much the infrastructure playing field has leveled. What strikes me is that Google's massive web index and PageRank algorithms might not be the moat they once were if the real value shifts to retrieval quality and LLM orchestration.
Perplexity’s innovative approach to real-time, cited answers truly sets a new standard for AI-powered search.
Is step 3 missing from the "how RAG works" diagram? Or am I not following it correctly?