惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

N
News and Events Feed by Topic
Malwarebytes
Malwarebytes
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
C
Cybersecurity and Infrastructure Security Agency CISA
F
Future of Privacy Forum
C
Cisco Blogs
T
The Exploit Database - CXSecurity.com
A
Arctic Wolf
S
Securelist
K
Kaspersky official blog
S
Schneier on Security
T
ThreatConnect
T
Tenable Blog
Spread Privacy
Spread Privacy
T
True Tiger Recordings
AWS News Blog
AWS News Blog
F
Fox-IT International blog
量子位
T
Threatpost
V
Vulnerabilities – Threatpost
C
CERT Recently Published Vulnerability Notes
Cisco Talos Blog
Cisco Talos Blog
GbyAI
GbyAI
宝玉的分享
宝玉的分享
腾讯CDC
G
Google Developers Blog
aimingoo的专栏
aimingoo的专栏
Cyberwarzone
Cyberwarzone
有赞技术团队
有赞技术团队
S
SegmentFault 最新的问题
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
V
Visual Studio Blog
U
Unit 42
雷峰网
雷峰网
cs.CV updates on arXiv.org
cs.CV updates on arXiv.org
Simon Willison's Weblog
Simon Willison's Weblog
O
OpenAI News
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
The GitHub Blog
The GitHub Blog
The Register - Security
The Register - Security
MyScale Blog
MyScale Blog
小众软件
小众软件
A
About on SuperTechFans
Last Week in AI
Last Week in AI
Y
Y Combinator Blog
博客园 - 三生石上(FineUI控件)
美团技术团队
Google Online Security Blog
Google Online Security Blog
P
Proofpoint News Feed
MongoDB | Blog
MongoDB | Blog

DEV Community

Qwen 3.6 Has Four Tiers. Here's How to Route Without Burning Cash. How I handle my errors in PHP The Blind Spot in Treasure Hunt Engine Configuration: Long-Term Server Health Run NVIDIA NIM on Your Own GPU — Same API, Different Endpoint Webflow SEO Implementation 로컬 LLM 셋업 가이드 (v21) How Logs Travel From Your EKS Pod to Datadog 𝗦𝘁𝗼𝗽 𝗖𝗿𝗮𝗺𝗺𝗶𝗻𝗴 𝗙𝗼𝗿 𝗘𝘅𝗮𝗺𝘀, 𝗦𝘁𝗮𝗿𝘁 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗥𝗲𝗮𝗹 𝗦𝗸𝗶𝗹𝗹𝘀 How to Use EXPLAIN ANALYZE in PostgreSQL: A Visual Guide gRPC Performance: tonic (Rust) vs grpc-go Benchmarked at Scale Hack The Box (HTB): Cap Machine (Full Walkthrough) Visual Search Optimization studygemma: AI study buddy for CS students Architectural Tradeoffs in Webhook Idempotency and SaaS API Versioning One Open Source Project a Day (No. 75): Understand Anything - The AI Engine That Turns Any Codebase Into an Explorable Knowledge Graph From mock-only-works to real-world-works: 48 hours of reCAPTCHA debugging I built a free music tool AI Talking Avatar Pipelines Broke Our Ad CTR by 3.7% 800G to 400G Breakout: How to Scale 400G Networks with 800G Ports 터미널 AI 에이전트 구축 (v20) Topical Authority Architecture Inside Hermes Agent's Session Memory: What X-Hermes-Session-Id Actually Does How Logs Travel From Your EKS Pod to Datadog The Hidden Journey Inside / Kubernetes Is it safe to connect my bank account to AI? No Room — The World of Aying (8/12) Fossils — The World of Aying (10/12) Familiar Stranger — The World of Aying (9/12) Being Seen — The World of Aying (7/12) [I Ran an AI Agent for 30 Days Straight — Here's the Boring Engineering That Made It Work] Gemma 4: The 128K Multimodal Powerhouse in Your Terminal How to Consolidate Your QA Toolstack: A Practical Buyer's Guide The Thank-You Email Almost Nobody Sends (And Why That's Your Edge) Schema Types 2026 Idempotency Keys: The API Safety Net You're Probably Not Using How to let Claude see my Plaid bank data Kiro Did It: Build a Simple Portfolio Website with Kiro IDE | From Prompt to HTML Prototype Islands of Commerce: What Marketplace Founders Can Learn from 60 Years of Island Biogeography React Pointer Hooks: Hover, Long-Press, Double-Click, Scratch, and Click-Outside Without the Bugs Engineering decisions for my video call tool VBScript Still Lives: How a Custom Go VM Brought Classic ASP to Linux and Mac What Happens When You Teach Old Scripting Languages New Runtime Tricks? I Tested 6 AI Coding Assistants for a Month. Here's What Actually Works. Extendscript Still Has Life Afriex Webhook Integration Guide: Signature Verification, Event Handling, and Production Best Practices The Blind Alleys of Veltrix Configuration How an ESP32 Turned a LEGO WALL-E Into a Real Working Robot The Flawed Promise of Real-Time Event Handling SSH Login Taking Forever? Check Your DNS Settings Found 897 Fake Followers on DEV.to. Here's How I Proved It. Retry logic, Kafka consumer lag, and the hidden failure pattern that Kubernetes won’t catch WebMCP Might Be the Most Important Announcement at Google I/O 2026 Build a Secure API with Rails 8 - Part-3: Auth Controllers I A/B tested 4 LLMs on the same 500 queries. The results surprised me. Google I/O 2026’s Smartest Developer Release Wasn’t a Model, It Was the Runtime - Managed Agents in Gemini API OSS Monthly Recap: What My Daily Commit Challenge Taught Me About Open Source “Culture” GemmaNotes Cognitive Debt: AI Is Building Your Systems. Do You Actually Understand Them? GeekNews Frontend Weekly Deep Dive - 2026-05-25 I Built a Universal Silicon Loader That Runs on Any SOC (No Bootrom Exploit) Docker容器化部署Node.js应用最佳实践 I Put a Neural Network in a Thermometer — Then It Got Out of Hand Building MGZon: Developer Portfolio + AI Bot + Social Network (9 min demo) Bearing Life (L10): What the Catalog Number Really Tells You Longhorn Volume Health: The Gap Between 'Healthy' and Actually Working Stop Prompting. Start Specifying: How Spec-Driven Development Fixes AI Coding TIL a PowerPoint file is just a zip — so I converted .pptx to Word entirely in the browser 로컬 LLM 셋업 가이드 (v18) Cx Dev Log — 2026-04-24 github's agent audit api is the boring feature that matters # From Teaching Code to Building Real-World Applications Vivado 2026.1 and Linux: why this decision matters beyond the headline Vivado 2026.1 y Linux: por qué la decisión importa más allá del titular ORA-00206 오류 원인과 해결 방법 완벽 가이드 Entidades finas e composição: o design que escolhi para a nova plataforma 10 Open Source Tools Every Developer Should Know 🔥 SSH Config File Mastery: Turning `~/.ssh/config` Into a Productivity Tool I tried to create a programming language... in python I Replaced 70MB Node.js Log Viewer with a 172KB Zig Binary I Turned npm outdated into a CI Gate — Here's How Don't fall for the Claude Mythos hype Vestige: A Gemma 4 Brain Tracker That Won't Blow Smoke Up Your Ass Gemminate: Transforming Static Textbooks into Interactive Learning Journeys with Gemma 4 Where Did All the Code Playgrounds Go? I built PROOFER - Privacy first Chrome extension that proofreads your texts using Gemma 4 I Automated My Entire Digital Product Business on a $13/Month GCP VM. Here's the Architecture. Beginner's Mind in Engineering and AI How I use AI agents to turn ideas into public demos I Built a Quotation Generator for Kenyan Street Welders Using Gemma 4's Vision The Math Behind Neural Networks — Explained Like Nobody Did for Me 🧨 Understanding TPC with IEEE802.11h What I’m Starting to Look for in Engineers An npm Downloads Comparison Chart in 300 Lines of Vanilla JS — Nice-Tick Math and API-Direct Fetch Vitreus: Local-First Spreadsheet Intelligence with Gemma 4 Transfer Fees, Metadata, and Soulbound Tokens: A Tour of Solana Token Extensions I got tired of re-explaining my codebase to ChatGPT — so I built a VS Code extension Revisiting My Phone AI After Gemma 4: The Upgrade I Didn't Know I Needed I built a privacy-first PDF merger in 7 hours — here's the stack and the lessons Google I/O 2026 made me ask an uncomfortable question: are we still coding, or are we managing builders? SSR with JavaScript: Escaping Node.js Clunkiness with AxonASP My CKA Exam-Day Experience: What Went Right, What Went Wrong, and Lessons Learned
RAG 시스템 실전 구축 (v21)
matias yoon · 2026-05-25 · via DEV Community

RAG 시스템 실전 구축 (v21)

1. RAG 기초 개념

Retrieval-Augmented Generation (RAG)은 검색 기반 생성 시스템으로, LLM의 지식 범위를 확장하는 데 효과적입니다. 핵심 루프는 다음과 같습니다:

입력 질의 → 검색기 → 문서 조각 → 증강된 프롬프트 → LLM 생성 → 출력

Enter fullscreen mode Exit fullscreen mode

RAG 시스템은 3가지 주요 구성 요소로 이루어집니다:

  1. Retriever: 관련 문서 찾기
  2. Generator: 증강된 프롬프트로 생성
  3. Embedding: 문서 및 질의 벡터화

2. 청킹 전략

2.1 의미적 청킹

from sentence_transformers import SentenceTransformer
from sklearn.cluster import KMeans
import numpy as np

class SemanticChunker:
    def __init__(self, model_name='all-MiniLM-L6-v2'):
        self.model = SentenceTransformer(model_name)

    def chunk_by_semantic(self, text, threshold=0.7):
        sentences = text.split('. ')
        embeddings = self.model.encode(sentences)

        # K-means 클러스터링으로 의미적 그룹 생성
        kmeans = KMeans(n_clusters=max(1, len(sentences)//3))
        kmeans.fit(embeddings)

        clusters = {}
        for i, cluster_id in enumerate(kmeans.labels_):
            if cluster_id not in clusters:
                clusters[cluster_id] = []
            clusters[cluster_id].append(sentences[i])

        return [' '.join(cluster) for cluster in clusters.values()]

Enter fullscreen mode Exit fullscreen mode

2.2 재귀적 청킹

import re

class RecursiveChunker:
    def __init__(self, chunk_size=512, overlap=50):
        self.chunk_size = chunk_size
        self.overlap = overlap

    def chunk_recursive(self, text):
        chunks = []
        start = 0

        while start < len(text):
            end = min(start + self.chunk_size, len(text))
            chunk = text[start:end]

            # 중간에 문장 분리점 찾기
            if end < len(text):
                separator = chunk.rfind('. ')
                if separator > self.chunk_size * 0.3:
                    end = start + separator + 1
                    chunk = text[start:end]

            chunks.append(chunk)
            start = max(0, end - self.overlap)

        return chunks

Enter fullscreen mode Exit fullscreen mode

3. 임베딩 모델 선택

3.1 모델 비교 테스트

from sentence_transformers import SentenceTransformer
import time

def compare_embedding_models():
    models = {
        'all-MiniLM-L6-v2': 'fast',
        'all-mpnet-base-v2': 'medium',
        'multi-qa-MiniLM-L6-v2': 'fast',
        'paraphrase-multilingual-MiniLM-v2': 'fast'
    }

    test_sentences = [
        "The quick brown fox jumps over the lazy dog.",
        "Machine learning is a subset of artificial intelligence.",
        "Natural language processing enables computers to understand text."
    ]

    results = {}
    for model_name, speed in models.items():
        model = SentenceTransformer(model_name)

        # 성능 측정
        start_time = time.time()
        embeddings = model.encode(test_sentences)
        end_time = time.time()

        results[model_name] = {
            'speed': end_time - start_time,
            'dimensions': len(embeddings[0]),
            'speed_category': speed
        }

    return results

# 사용 예시
model_results = compare_embedding_models()
print("모델 성능 비교:")
for name, metrics in model_results.items():
    print(f"{name}: {metrics['speed']:.2f}s, {metrics['dimensions']}차원")

Enter fullscreen mode Exit fullscreen mode

4. 벡터 데이터베이스 비교

4.1 Chroma vs Qdrant vs pgvector

import chromadb
from qdrant_client import QdrantClient
import psycopg2
import numpy as np

class VectorDBComparison:
    def __init__(self):
        # Chroma
        self.chroma_client = chromadb.Client()
        self.chroma_collection = self.chroma_client.get_or_create_collection(
            name="rag_test"
        )

        # Qdrant
        self.qdrant_client = QdrantClient(host='localhost', port=6333)

        # pgvector (PostgreSQL)
        self.pg_conn = psycopg2.connect(
            host="localhost", 
            database="rag_db", 
            user="user", 
            password="password"
        )

    def chroma_insert(self, documents, embeddings):
        self.chroma_collection.add(
            documents=documents,
            embeddings=embeddings,
            ids=[f"doc_{i}" for i in range(len(documents))]
        )

    def qdrant_insert(self, documents, embeddings):
        self.qdrant_client.upsert(
            collection_name="rag_test",
            points=[
                {
                    "id": i,
                    "vector": emb.tolist(),
                    "payload": {"text": doc}
                }
                for i, (doc, emb) in enumerate(zip(documents, embeddings))
            ]
        )

    def pgvector_insert(self, documents, embeddings):
        with self.pg_conn.cursor() as cursor:
            for i, (doc, emb) in enumerate(zip(documents, embeddings)):
                cursor.execute(
                    "INSERT INTO documents (id, content, embedding) VALUES (%s, %s, %s)",
                    (i, doc, np.array(emb).tobytes())
                )
            self.pg_conn.commit()

Enter fullscreen mode Exit fullscreen mode

5. 전체 RAG 파이프라인 코드

import chromadb
from sentence_transformers import SentenceTransformer
from typing import List, Dict
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity

class SimpleRAGPipeline:
    def __init__(self, chunk_size: int = 512):
        self.chunker = RecursiveChunker(chunk_size=chunk_size)
        self.embedder = SentenceTransformer('all-MiniLM-L6-v2')
        self.chroma_client = chromadb.Client()
        self.collection = self.chroma_client.get_or_create_collection(
            name="documents"
        )

    def add_documents(self, documents: List[str]):
        """문서 추가 및 임베딩"""
        all_chunks = []
        all_texts = []

        # 문서 청킹
        for doc in documents:
            chunks = self.chunker.chunk_recursive(doc)
            all_chunks.extend(chunks)
            all_texts.extend([doc] * len(chunks))

        # 임베딩 생성
        embeddings = self.embedder.encode(all_chunks)

        # Chroma에 저장
        self.collection.add(
            documents=all_chunks,
            embeddings=embeddings.tolist(),
            ids=[f"chunk_{i}" for i in range(len(all_chunks))]
        )

    def retrieve(self, query: str, top_k: int = 3) -> List[Dict]:
        """검색"""
        query_embedding = self.embedder.encode([query])
        results = self.collection.query(
            query_embeddings=query_embedding.tolist(),
            n_results=top_k
        )

        return [
            {
                'text': text,
                'score': score
            }
            for text, score in zip(results['documents'][0], results['distances'][0])
        ]

    def generate_response(self, query: str, retrieved_docs: List[Dict]) -> str:
        """응답 생성"""
        context = "\n".join([doc['text'] for doc in retrieved_docs])
        prompt = f"""
        주어진 컨텍스트를 기반으로 질문에 답변해주세요.

        컨텍스트:
        {context}

        질문: {query}
        답변:
        """

        # 간단한 LLM 호출 예시 (실제 구현에서는 실제 LLM 사용)
        return f"검색된 컨텍스트 기반 답변: {query}에 대한 정보는 {len(retrieved_docs)}개 문서에 포함되어 있습니다."

# 사용 예시
rag = SimpleRAGPipeline()

# 문서 추가
documents = [
    "Python은 고급 프로그래밍 언어로, 간결하고 읽기 쉬운 문법을 갖추고 있습니다.",
    "Django는 파이썬 기반 웹 프레임워크로, 빠르고 안전한 웹 개발을 지원합니다.",
    "React는 JavaScript 라이브러리로, 사용자 인터페이스 개발에 최적화되어 있습니다."
]

rag.add_documents(documents)

# 검색 및 응답 생성
query = "Python 프로그래밍 언어의 특징은?"
retrieved = rag.retrieve(query)
response = rag.generate_response(query, retrieved)
print(f"질의: {query}")
print(f"응답: {response}")

Enter fullscreen mode Exit fullscreen mode

6. 고급 기능

6.1 질의 변환


python
class QueryTransformer:
    def __init__(self):
        self.embedder = SentenceTransformer('all-MiniLM-L6-v2')

    def transform_query(self, original_query: str) -> List[str]:
        """질의 변환

---

📥 **Get the full guide on Gumroad**: https://gumroad.com/l/auto ($7)

Enter fullscreen mode Exit fullscreen mode