Как спроектировать AI ассистента для поддержки и не слить бюджет. Часть 1: От идеи до выбора архитектуры

Каждая команда сейчас хочет заменить людей на AI. Но есть другой подход - усилить текущие возможности с помощью AI. Это может быть крутым драйвером для масштабирования и роста, а не поводом для увольнения. Покажу как спроектировать агента который решает проблемы и можно внедрять в продакшен. Буду рассказывать на примере юридической поддержки, но подход универсальный. Поговорим про RAG и GraphRAG, про развёртывание и выбор модели. Статья будет полезна как для больших компаний так и для маленьких - разница только в том где вы будете хранить и обрабатывать данные с LLM.

https://habr.com/ru/articles/975230/

#RAG #graphrag #ai #llmагент #ииассистент #автоматизация_поддержки

📣 3x 6-month #internships at Wimmics for M2 students! Work on #NeuroSymbolicAI at the intersection of #MachineLearning & #KnowledgeGraphs: analogical reasoning, KG tasks, structured memory, #RAG, #GraphRAG, #LLM. Co-supervision:
@fabien_gandon
& Ndeye-Emilie Mbengue.
📍 Where: Sophia Antipolis (France)
⏳ Duration: 6 months
Details below!

Advanced RAG: LongRAG, Self-RAG and GraphRAG Explained:
https://www.glukhov.org/post/2025/11/advanced-rag-variants-longrag-self-rag-graphrag/
#RAG #longrag #selfrag #graphrag #python #ai #llm

FalkorDB’s GraphRAG just outperformed traditional vector retrieval by 3.4× on structured benchmarks. By blending LLM reasoning with schema‑aware graph constraints, it delivers faster, more accurate results for hybrid data pipelines. Curious how hybrid approaches and metadata handling give this edge? Dive into the details. #FalkorDB #GraphRAG #LLM #VectorRetrieval

🔗 https://aidailypost.com/news/falkordb-graphrag-beats-vector-retrieval-34-structured-benchmarks

Graph Retrieval-Augmented Generation: A Survey. A bit heavy on the property graph and less on the RDF side of things, but fun reading for the connaisseurs out there. https://github.com/pengboci/GraphRAG-Survey #GraphRAG #KnowledgeGraphs

Last week we hosted #PyDataVenice #23 🎉

🕸️ We kicked off with a hands-on session on the basics of #GraphRAG
📈 Then we dived into exploring LLMs in #Finance

Huge thanks to everyone who made this evening possible! 🙌

🎦 If you missed the #meetup, no worries: you can watch the #livestream on #YouTube:
👉 https://www.youtube.com/watch?v=_1ApTJuY-68

💾 See you all on Thursday, December 18 #SaveTheDate! 💪

#PyDataVE #23 #Meetup #PyData - #AI #LLM #RAG

온톨로지 전문가가 사라진다: LLM이 바꾸는 지식 그래프의 미래

https://aisparkup.com/posts/5918

SuperMemory is just one example of a growing ecosystem of knowledge graph systems (Graphiti- Zep, Fast Graph RAG, TrustGraph...) Some in Python, some TypeScript with the added benefit of having graph visualization. Even in Rust and Go there is a growing list of open source graph-RAG.
https://github.com/supermemoryai #KnowledgeGraph #GraphRAG

CodeGraphContext – server MCP tạo đồ thị mã nguồn, cung cấp ngữ cảnh chính xác cho AI coding (Gemini, Copilot, Cursor…). Tự động xây dựng bản đồ kiến trúc, đồng bộ tài liệu, hỗ trợ navigation, completions, debugging. Đã có package Python, website, repo GitHub và cộng đồng Discord 50 dev. #AI #Coding #GraphRAG #OpenSource #CodeGraphContext #MãNguồn #TríTuệNhânTạo

https://www.reddit.com/r/programming/comments/1o2ezeu/i_built_codegraphcontext_an_mcp_server_that/

The next meetup will include talks in Italian 🇮🇹 and in English 🇬🇧

🕖 We’ll meet #InPresence on Thursday, October 30 at 7:00 PM.
🏠 We’ll be hosted by Anda Venice Hostel in the Medusa Room and, for those who cannot join us, it will also be on YouTube in #streaming.

👥 And afterwards, for those who want, we’ll stay there for ( 🍸 | 🍺 ) & ( 🍕 | 🍔 ) and #networking until late ‼️

Info & Reservations 👇
https://www.meetup.com/pydata-venice/events/305865171/

#PyDataVE #23 #Meetup #PyData - #GraphRAG #Analysis #Forecasting

🌘 ApeRAG：具備多模態索引、AI 代理、MCP 支援及可擴展 K8s 部署的生產級 GraphRAG 平臺
➤ 解鎖智慧應用的生產級 GraphRAG 解決方案
✤ https://github.com/apecloud/ApeRAG
ApeRAG 是一個專為生產環境設計的檢索增強生成（RAG）平臺，它整合了圖譜 RAG、向量搜尋及全文檢索，並輔以先進的 AI 代理。此平臺支援多模態文件處理、智慧代理，並提供企業級管理功能，是建構知識圖譜、情境工程及部署自主搜尋推理 AI 代理的理想選擇。ApeRAG 透過 Docker Compose 或 Kubernetes 進行部署，並支援 Model Context Protocol (MCP) 讓 AI 助理直接與知識庫互動。此外，它還整合了 MinerU 技術以提供進階的文件解析功能，並包含向量、全文、圖譜、摘要及視覺等多種索引類型。
+ 這個平臺的部署方式聽起來很直觀，Docker Compose 讓開發者能快速上手，K8s 部署也為大規模應用提供了保障。

#AI #RAG #GraphRAG #Kubernetes #多模態

GraphRAG로 AI가 드디어 ‘진짜 똑똑해졌다’ – 기업들이 열광하는 이유

https://aisparkup.com/posts/4652

cogito:14b has become my go-to #localLLM for creating structured JSON output, like for #GraphRAG. Of the models that will fit in < 16GB VRAM, this is the one that consistently gets the job done. #ollama #n8n

Memgraph v3.4 is out and, among much more, you can now have vectors on edges, quantize vector indexes, hierarchical graph layout, Ollama integration...The graph chat in the Lab is so much fun. Well done guys, totally fab. https://memgraph.com/memgraphdb
#GraphDatabase #GraphRAG

PuppyGraph is a delightful intermediate between your relational and NoSQL data. You can use Cypher or Gremlin, no ETL. Claims 6-hop traversals below 1s. Also boasts #GraphRAG atop MongoDB. https://www.puppygraph.com/ #GraphDatabase

I think #graphrag with LLMs is too expensive for private use. Especially for documents tasks with #ragflow.

The token output costs of Gemini are too high, and it's among the cheapest these days.

DeepSeek still has an edge here. But it doesn't perform well for English imho.

#RAGflow is a very interesting knowledge management engine for #GenAI workflows.

https://because-security.atlassian.net/wiki/spaces/LML/pages/621641933/RAGflow+local+MCP+server+to+connect+Claude+Desktop+to+your+expert+Knowledge+Base

Many use cases related to Context Engineering to provide the data to a LLM / tool to process it.

#DeepDoc, #GraphRAG with OCR models (Mistral or other models). Cutting-edge stuff.

Sadly, the #Anthropic API costs for output tokens are expensive, but Chat models from Google can fill this gap.

🎉 Thrilled to announce Roberto Navigli as Keynote Speaker at #RAGEKG 2025! 🎤 Full Professor at the Sapienza University of Rome, founder of Babelscape, and pioneer #GraphRAG research!

Nov 2-3 | Nara, Japan 🇯🇵
🔗 Details and CfP (open): https://2025.rage-kg.org

@iswc_conf @ragekg #AI #KnowledgeGraphs #LLM

Impressive multilingual and multimodal embedding model: Jina Embeddings v4.
#LLMs #GraphRAG
https://jina.ai/news/jina-embeddings-v4-universal-embeddings-for-multimodal-multilingual-retrieval/

Использование графов знаний при разработке RAG-систем

Привет, Habr! На связи Александр Сулейкин, Роман Бабенко и Даниил Бутнев. Подготовили совместную статью по теме использования графов знаний при разработке RAG-систем. В рамках статьи рассказываем про основные проблемы традиционных RAG-систем, даем обзор основных открытых проектов графов знаний GraphRAG, показываем и даем краткое описание архитектуры таких систем, а также рассказываем про практическое использование графов знаний на примере трех областей - медицины, метрологии и стандартизации.

https://habr.com/ru/articles/908890/

#graphrag #LightRAG #RAG #ai #llm