Piyush Garg

Hey Everyone, I am excited to announce teachyst.com 🎉 A whole new platform for Instructors and Content Creators. 🥳 🧵 Thread

@largedietcoke_ @mannupaaji 😂😂😂😂😂😂😂😂

Babbu Maan te Karan Aujla da collab

Class ke pehle ek live to bnta tha aur sir aa gya @ChaiCodeHQ @piyushgarg_dev

Class ke time par video release kariii😭 Par it is such a pleasure to watch my 2 fav techies together @piyushgarg_dev @mannupaaji

I sat down with @piyushgarg_dev and discussed how to get a girlfriend in this fast moving AI world Do watch!

@mannupaaji Just in case you are wondering why am I single? because coaches don’t play 😁

Key takeaways from the video @piyushgarg_dev Why are vector RAG dead and what are vectorless RAG? When you send PDFs to a LLM and based on it get your queries answered Now if you give a lot of data 1) quality detoriates give general answers 2) a high cost of tokens RAG solved this problem with two phases 1) indexing : gives files and do chunking, divide files into chunks Call embedding models and get array of numbers stored in vector db 2) query: using same vector embedding model, make them an array of vectors. Now search for similar number in database using vector similarity search Now using LLM on the relevant numbers perform a generation of output Problem with vector rag 1) When you do window fixed chunking i.e. divide into 500 words window then you might have lost context of the file 2) legal docs have reference ex: as per rule 63.7.4 of A so now we want reference of that page too 3) User might send any query words, they might not give the exact same word so the relevant vector embeddings might not get generated properly Vectorless RAG works on a tree structure i.e. hierarchical table of contents Page index follows structure Document -> hierarchical index -> reasoning based retrieval and finally output It relies on narrative structure and not fixed chunk size Now based on user query instead of giving full script LLM uses the tree and find the most relevant nodes, now find original chunks of that node and fetch the output Point to remember : vectorless RAG come with their own negatives i.e. speed and cost of LLMs Blog referred: @visrow/what-is-pageindex-how-to-build-a-vectorless-rag-system-no-embeddings-no-vector-db-dc097fae3071" target="_blank" rel="nofollow noopener">medium.com/@visrow/what-i…

Skill sikhte hi dead ho jaata hai 😑

answer to DSA vs DEV

@isudiptodas Thanks a lot man! Means a lot ❤️❤️

Probably the only person in my youtube whose video I never summarized, instead I chose to listen every minute of the OG @piyushgarg_dev 🙌🏻

It's is a fully knowledge pack video @piyushgarg_dev

most underrated channel on YouTube @piyushgarg_dev for software engineering 🙌🙌

Yesterday was another banger session by Piyush Garg on consistent hashing 🔥 No regrets skipping Dhurandhar 2 first day for this one. Got introduced to some solid concepts: Eviction policies Magic numbers , CacheDBHotspots Sessions that levels up your system design knowledge

Made this video back in 2023 when LLMs was not a hot topic and no one was talking about RAG 😀. Still a relevant video, Do check it out on my YouTube channel.

Learning in public 🔥 Thanks @piyushgarg_dev sir for this video this is really helps a lot Built a Beginner Guide for Hosting Node.js on Dedicated VPS 🎥 Video: youtu.be/YGW_uRefvkQ?si… @ChaiCodeHQ @Hiteshdotcom

Day 7 with @piyushgarg_dev ​Distributed Caching ​Designing systems like Redis from scratch. Consistent Hashing for seamless scaling. ​Virtual Nodes, hash points per server perfectly balanced data distribution. ​#SystemDesignChaiCode #Redis #Backend #Caching #DistributedSystems

Learned today about Consistent hashing and the problems faced in Consistent hashing.

Learned how real systems handle scale today: Designed a cache store Understood TTL and eviction policies Explored distributed caching Then came the interesting part : consistent hashing Why we need it, and the problems it still brings @piyushgarg_dev

Amazing session today . - Learnt about cache , - designing a cache store , - from choice of db for cache to consistent hashing and virtual nodes - most importantly , acting tips from the legend @piyushgarg_dev sir ! 🙌🏻

System Design Session 7 ✅ Started from Caching -> Designed and Scaled Caching System (Redis) -> Eviction Policy (LRU) -> Traditional Hashing VS Consistent Hashing (Less Data Movement) Eagerly waiting for the Next One (IYKYK)✌️

Attend the today's Class: Understand about Caching, Distributed Caching and Consistent Hashing. Aaj @piyushgarg_dev acting tips bhi mili. @piyushgarg_dev @Hiteshdotcom @nirudhuuu @ChaiCodeHQ