Post

๐Ÿ”Š Create your own LLM RAG application in just 3 days using this hands-on roadmap crafted from the best free resources!

Its appeal lies in its lightweight design and the simplicity of integrating it with any foundational LLM.

๐Ÿ’ก Use this 3-day guide to step into the evolving landscape of RAG and its latest developments! Spend 2-3 hours every day on the resources.

๐Ÿฅ Youโ€™ll start with the basics, move to advanced ideas, build your app using LangChain and learn how to evaluate it. Iโ€™ve also added resources to catch up on the latest research in the space.

The roadmap with resources: https://github.com/aishwaryanr/awesome-generative-ai-guide/blob/main/resources/RAG_roadmap.md

โ›ณ Day 1: Introduction to RAG

  • ๐Ÿ‘‰ What is Retrieval Augmented Generation?
  • ๐Ÿ‘‰ Key components of RAG: Ingestion, Retrieval, Synthesis
  • ๐Ÿ‘‰ RAG pipeline components: Chunking, Embedding, Indexing, Top-k Retrieval and Generation

โ›ณ Day 2: Advanced RAG + Build Your Own RAG System

  • ๐Ÿ‘‰ Optimizations for Advanced RAG: Self Querying Retrieval, Parent Document ๐Ÿ‘‰ Retriever, Hybrid Search, Compressors, HyDE etc.
  • ๐Ÿ‘‰ Build your own RAG system with LangChain and OpenAI Resources for building advanced RAG applications

โ›ณDay 3: RAG Evaluation and Challenges

  • ๐Ÿ‘‰Commonly used evaluation metrics from TruEra and RAGas
  • ๐Ÿ‘‰RAG Pain Points and Solutions

๐Ÿฅ๐ŸฅThe roadmap also includes: Optional Reading Resources & Top 2024 RAG research papers

 3 Day RAG Roadmap

Translate to Korean

RAG(Retrieval Augmented Generation)๋Š” LLM ๋ถ„์•ผ์—์„œ ๋งค์šฐ ์ธ๊ธฐ ์žˆ๋Š” ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์œผ๋กœ ๋ถ€์ƒํ–ˆ์Šต๋‹ˆ๋‹ค.

์ด ๊ฒŒ์ž„์˜ ๋งค๋ ฅ์€ ๊ฒฝ๋Ÿ‰ ์„ค๊ณ„์™€ ๊ธฐ๋ณธ LLM๊ณผ ํ†ตํ•ฉํ•  ์ˆ˜ ์—†๋‹ค๋Š” ์ ์— ์žˆ์Šต๋‹ˆ๋‹ค.

๐Ÿ’ก ์ด 3์ผ ๊ฐ€์ด๋“œ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ RAG์˜ ์ง„ํ™”ํ•˜๋Š” ํ™˜๊ฒฝ๊ณผ ์ตœ์‹  ๊ฐœ๋ฐœ์— ๋Œ€ํ•ด ์•Œ์•„๋ณด์„ธ์š”! ๋งค์ผ 2-3 ์‹œ๊ฐ„์„ ์ž์›์— ํˆฌ์žํ•˜์‹ญ์‹œ์˜ค.

๐Ÿฅ ๊ธฐ๋ณธ ์‚ฌํ•ญ๋ถ€ํ„ฐ ์‹œ์ž‘ํ•˜์—ฌ ๊ณ ๊ธ‰ ์•„์ด๋””์–ด๋กœ ์ด๋™ํ•˜๊ณ , LangChain์„ ์‚ฌ์šฉํ•˜์—ฌ ์•ฑ์„ ๋นŒ๋“œํ•˜๊ณ , ํ‰๊ฐ€ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋ฐฐ์›๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด ๋ถ„์•ผ์˜ ์ตœ์‹  ์—ฐ๊ตฌ๋ฅผ ๋”ฐ๋ผ์žก์„ ์ˆ˜ ์žˆ๋Š” ๋ฆฌ์†Œ์Šค๋„ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.

โ›ณ 1์ผ์ฐจ: RAG ์†Œ๊ฐœ

  • ๐Ÿ‘‰ ๊ฒ€์ƒ‰ ์ฆ๊ฐ• ์ƒ์„ฑ์ด๋ž€ ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?
  • ๐Ÿ‘‰ RAG์˜ ํ•ต์‹ฌ ๊ตฌ์„ฑ ์š”์†Œ: Ingestion, Retrieval, Synthesis
  • ๐Ÿ‘‰ RAG ํŒŒ์ดํ”„๋ผ์ธ ๊ตฌ์„ฑ ์š”์†Œ: ์ฒญํฌ, ์ž„๋ฒ ๋”ฉ, ์ธ๋ฑ์‹ฑ, Top-k ๊ฒ€์ƒ‰ ๋ฐ ์ƒ์„ฑ

โ›ณ 2์ผ์ฐจ: ๊ณ ๊ธ‰ RAG + ๋‚˜๋งŒ์˜ RAG ์‹œ์Šคํ…œ ๊ตฌ์ถ•ํ•˜๊ธฐ

  • ๐Ÿ‘‰ ๊ณ ๊ธ‰ RAG ์ตœ์ ํ™”: Self Querying Retrieval, Parent Document ๐Ÿ‘‰ Retriever, Hybrid Search, Compressors, HyDE ๋“ฑ
  • ๐Ÿ‘‰ LangChain ๋ฐ OpenAI๋กœ ์ž์ฒด RAG ์‹œ์Šคํ…œ ๊ตฌ์ถ• ๊ณ ๊ธ‰ RAG ์‘์šฉ ํ”„๋กœ๊ทธ๋žจ ๊ตฌ์ถ•์„ ์œ„ํ•œ ๋ฆฌ์†Œ์Šค

โ›ณ3์ผ์ฐจ: RAG ํ‰๊ฐ€ ๋ฐ ๊ณผ์ œ

  • ๐Ÿ‘‰TruEra ๋ฐ RAGas์—์„œ ์ผ๋ฐ˜์ ์œผ๋กœ ์‚ฌ์šฉ๋˜๋Š” ํ‰๊ฐ€ ์ง€ํ‘œ
  • ๐Ÿ‘‰RAG์˜ ๋ฌธ์ œ์  ๋ฐ ํ•ด๊ฒฐ ๋ฐฉ๋ฒ•

๐Ÿฅ๐Ÿฅ๋กœ๋“œ๋งต์—๋Š” ๋‹ค์Œ ๋‚ด์šฉ๋„ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค: Optional Reading Resources & Top 2024 RAG research papers

This post is licensed under CC BY 4.0 by the author.