Back
Rahul Agarwal
Founder | Agentic AI... â˘Â 1m
Most people don't even know these basics of RAG. I've explained it in a simple way below. 1. đđťđąđ˛đ đśđťđ´ Convert documents into a format that AI can quickly search later. Step-by-step: ⢠đđźđ°đđşđ˛đťđ: You start with files like PDFs, Word docs, notes, websites, etc. ⢠đđ đđżđŽđ°đ â đ§đ˛đ đ: The system pulls raw text out of those documents. ⢠đđľđđťđ¸đ: The long text is broken into đđşđŽđšđš đ˝đśđ˛đ°đ˛đ (chunks). This is important because AI canât understand very large text at once. ⢠đŠđ˛đ°đđźđżđśđđ˛ / đđťđ°đźđąđ˛: Each chunk is converted into numbers called đđ˛đ°đđźđżđ. These numbers represent the đŽđŚđ˘đŻđŞđŻđ¨ of the text. ⢠đđşđŻđ˛đąđąđśđťđ´ đ đźđąđ˛đš: A special model does this text â vector conversion. ⢠đŚđŽđđ˛ đśđť đŠđ˛đ°đđźđż đđŽđđŽđŻđŽđđ˛: All vectors are stored in a đđ˛đ°đđźđż đąđŽđđŽđŻđŽđđ˛ so they can be searched later. ________________ 2. đĽđ˛đđżđśđ˛đđŽđš (đĽ) Fetch the most relevant chunks for a userâs question. Step-by-step: ⢠đ¨đđ˛đż đđđŻđşđśđđ đŽ đžđđ˛đđđśđźđť: Example: âđđŠđ˘đľ đĽđ°đŚđ´ đľđŠđŚ đ¤đ°đŻđľđłđ˘đ¤đľ đ´đ˘đş đ˘đŁđ°đśđľ đľđŚđłđŽđŞđŻđ˘đľđŞđ°đŻ?â ⢠đ¤đđ˛đđđśđźđť â đŠđ˛đ°đđźđżđśđđ˛đą: The question is also converted into a vector using the same embedding engine. ⢠đŠđ˛đ°đđźđż đđŽđđŽđŻđŽđđ˛ đŚđ˛đŽđżđ°đľ: The system compares: 1. Question vector 2. Stored document vectors ⢠đ đŽđđ°đľđśđťđ´ / đŚđśđşđśđšđŽđżđśđđ đŚđ˛đŽđżđ°đľ: The database finds chunks whose meaning is closest to the question. ⢠đđ˝đ˝đżđźđ˝đżđśđŽđđ˛ đđľđđťđ¸đ đ˘đđđ˝đđ: Only the đşđźđđ đżđ˛đšđ˛đđŽđťđ đ˝đśđ˛đ°đ˛đ đźđł đđ˛đ đ are returned. ________________ 3. đđđ´đşđ˛đťđđŽđđśđźđť (đ) Enhance the userâs question with relevant information. Step-by-step: ⢠đĽđ˛đšđ˛đđŽđťđ đđľđđťđ¸đ: The retrieved text pieces are collected. ⢠đ đ˛đżđ´đ˛ đđśđđľ đŚđźđđżđ°đ˛ đđźđťđđ˛đťđ: These chunks are combined into a clean context block. ⢠đŁđżđźđşđ˝đ đđżđ˛đŽđđśđźđť: The system builds a new prompt: 1. Userâs original question 2. Retrieved context ⢠đđđ´đşđ˛đťđ đđľđ˛ đŁđżđźđşđ˝đ: This enriched prompt gives the AI đŁđ˘đ¤đŹđ¨đłđ°đśđŻđĽ đŹđŻđ°đ¸đđŚđĽđ¨đŚ. ________________ 4. đđ˛đťđ˛đżđŽđđśđźđť (đ) Generate a correct, grounded response. Step-by-step: ⢠đđťđżđśđ°đľđ˛đą đŁđżđźđşđ˝đ đŚđ˛đťđ: The prompt (question + context) is sent to the LLM. ⢠đđđ đ đźđąđ˛đšđ (đ˘đ˝đ˛đťđđ / đźđđľđ˛đżđ): The language model reads: 1. The question 2. The retrieved knowledge ⢠đđśđťđŽđš đ˘đđđ˝đđ: The model generates a response đŻđŽđđ˛đą đźđť đ˝đżđźđđśđąđ˛đą đ°đźđťđđ˛đ đ, not guesses. Why RAG Is Powerful? <> Normal LLMs rely only on training data BUT, <> RAG lets LLMs use đđźđđż đ˝đżđśđđŽđđ˛ đźđż đłđżđ˛đđľ đąđŽđđŽ and it's easy to update knowledge anytime. â Repost for others so they can understand the very basics of RAG.
Replies (4)
More like this
Recommendations from Medial
Kimiko
Startups | AI | info... â˘Â 8m
Vector databases for AI memory just got disrupted⌠by MP4 files?! Video as Database: Store millions of text chunks in a single MP4 file Store millions of text chunks with blazing-fast semantic search â no database required. 100% open source. Zero
See More
Rahul Agarwal
Founder | Agentic AI... â˘Â 14d
6 Chunking Methods for RAG you should know. Iâve explained it in a simple, step by step way. đŞđľđŽđ đśđ đđľđđťđ¸đśđťđ´? 1. Chunking means splitting large documents into smaller pieces. 2. Helps LLMs search and understand data better. 3. Essent
See More
Sushant Kumar
 â˘Â
VIVID MYND â˘Â 9m
Internship Alert â Vivid Mynd Technologies We're on the hunt for RAG-savvy interns! If you're into LlamaIndex, RAG concepts, and vector DBs like Chroma, this is your shot. Duration: 2 months Stipend: Fixed Who: College students or freelancers with
See MoreRahul Agarwal
Founder | Agentic AI... â˘Â 13d
Prompt vs Context vs RAG. I've explained it in a simple way below. 1. đŁđżđźđşđ˝đ đđťđ´đśđťđ˛đ˛đżđśđťđ´ Prompt Engineering is about đ°đšđ˛đŽđż đśđťđđđżđđ°đđśđźđťđ, not magic words. ⢠đđ˛đłđśđťđ˛ đđľđ˛ đłđśđťđŽđš đ´đźđŽđš: What exactly do
See More
Rahul Agarwal
Founder | Agentic AI... â˘Â 2m
Turn your laptop into a powerful RAG engine. The worldâs smallest vector database. I've explained below. If youâre building RAG today, youâre facing a brutal reality: đđšđźđđą đđ˛đ°đđźđż đđđ đŽđżđ˛ đ˛đ đ˝đ˛đťđđśđđ˛ and đźđť-đ˝đżđ˛đşđśđđ˛
See More
Krish Jaiman
Passionate about tec... â˘Â 1y
AI is reaching heights these days.But have you ever wondered how ChatGPT answers the question from recent affairs? Because see ChatGPT trained on the news of 31st January 2025 but it still answers the question relating to it .How? This is because of
See More
Saswata Kumar Dash
Buidling FedUp| AI R... â˘Â 8m
đ¨ Everyone says "RAG is dead" â but I say: Itâs just been badly implemented. Iâve worked on AI systems where Retrieval-Augmented Generation (RAG) either changed the game⌠or completely flopped. Hereâs the hard truth đ --- 𤯠Most teams mess up
See More
Chirotpal Das
Building an AI eco-s... â˘Â 10m
I feel pride in announcing our Made in India, Real-Time Vector Database - SwarnDB. SwarnDB is a SarthiAI initiative towards an effort to create an end to end eco-system for the future of AI. We tested SwarnDB with 100K vector records of 1536 dime
See MoreRahul Agarwal
Founder | Agentic AI... â˘Â 2m
2 ways AI systems today generate smarter answers. Iâve explained both in simple steps below. đĽđđ (đĽđ˛đđżđśđ˛đđŽđš-đđđ´đşđ˛đťđđ˛đą đđ˛đťđ˛đżđŽđđśđźđť) (đ´đľđŚđą-đŁđş-đ´đľđŚđą) RAG lets AI fetch and use real-time external information to ge
See More
/entrackr/media/post_attachments/wp-content/uploads/2021/08/Accel-1.jpg)
Download the medial app to read full posts, comements and news.