rag

A HuggingFace article summarizing a paper by IIT Madras researchers that groups LLM memory handling into five approaches — store more, compress, correct, learn at inference, and distribute plasticity — and questions how they compare when new information conflicts with old state.

A discussion of an often-overlooked agent evaluation case: what happens when a user's data access is revoked mid-conversation, covering stale context, cached retrievals, and cross-account cache leaks, and where enforcement should live in the stack.

The post explains why RAG needs both embedding-based retrieval and reranking: dual-encoder models are fast and allow offline precomputation of document vectors, making them suitable for large-scale retrieval but weaker at capturing fine details; cross-encoders score each pair more accurately but are computationally expensive, so recall-then-rerank is a necessary trade-off.

The author shares a learning roadmap for RAG and memory systems geared toward Agent development and interviews, covering chunking, hybrid retrieval, reranking, vector database selection, and short-term/long-term memory management, along with reference materials including the original RAG paper by Lewis et al.

The author explores how RAG evaluations should distinguish three different reasons a system answers "I don't know" — retrieval failure, missing evidence, and conflicting sources — and proposes evals that label the expected unresolved state rather than just scoring answers as correct, incorrect, or refused.

swyx argues that data engineers building data-intensive applications are in far higher demand than typical "RAG guy" AI engineers, and is recruiting for a mission-focused role via Latent Space, with relocation and visa support offered.

An in-depth article on a high-frequency Agent interview question — "Walk me through your RAG pipeline" — systematically covering the principles, trade-offs, and common follow-up questions across the full RAG workflow: offline indexing (parsing, chunking, vectorization, index building) and online QA (retrieval, reranking, context construction).

Qwen's 10-million-document RAG pipeline uses four lines of defense — hybrid retrieval, citation-constrained generation, claim-level verification, and calibrated abstention — to bring the hallucination rate on unanswerable questions down to 2%, while keeping p95 latency at 18ms. It's a production-grade anti-hallucination architecture worth studying.

该论文研究不确定性度量何时能有效衡量 RAG 幻觉风险，发现当检索证据充分约束答案时，困惑度能达到 0.81–0.91 AUROC，而检索覆盖不足时同一指标会失效，提出不确定性是条件性的幻觉风险度量。

SAG2 提出一种无需嵌入的多跳检索架构，仅使用 SQLite、BM25 和 JEV 上下文证据决策，在 MuSiQue、2WikiMultiHopQA 和 HotpotQA 上取得最高 Recall@5/Recall@10，索引速度快且无需模型调用。

DeepSeek 推出 Search API 与 Fetch API，为智能体和 RAG 应用提供实时网页内容访问能力，成功请求定价为每 1000 次 $0.75。Search API 由 DeepSeek 自研搜索引擎驱动，返回结构化结果；Fetch API 可将单个网页内容以 Markdown 格式返回。

Odin is a browser-based debugging search engine that parses a repo and ranks the exact file (93%) and function (79.4%) to fix for a given bug, using Tree-sitter via WebAssembly and sub-1B local models via WebGPU, matching Claude 4.5 agent accuracy at a fraction of the cost on SWE-bench Lite.

A team shares how their self-hosted RAG support chatbot confidently blended two warranty policies and promised a customer a free repair, and compares options for catching such errors — tracing retrieved chunks with tools like Langfuse or Arize Phoenix, versus blocking unchecked answers with guardrails from Orq AI or NeMo Guardrails.

作者预告将于周一/周二发布一篇关于 AI Engineering 的长文，涵盖上下文工程、RAG、Agent 编排、评估、推理服务等 20 个核心主题，已投入约 10 天深度调研并收集 112 个资源。

llmwiki 是一个开源的“知识编译器”，可将论文、笔记、PDF、网页等原始材料编译成带引用溯源、审查状态和检索元数据的互链 Markdown wiki，供 agent 和人类复用。它支持混合检索（语义搜索、BM25 重排、wikilink 图扩展），并提供 CLI、本地查看器、MCP server 和 TypeScript SDK 多种访问方式。

A developer shares a production-grade RAG assistant built with Python/FastAPI featuring hybrid retrieval (pgvector + Postgres FTS fused via RRF), parent-child chunking, cross-encoder reranking, and Promptfoo-based evals, and is seeking code review and architecture feedback.

Galileo has released the Mastering GenAI Series for free — four ebooks totaling around 540 pages, systematically covering hands-on engineering knowledge in AI Agent design, RAG, multi-agent architectures, and LLM-as-a-Judge evaluation. Ideal for developers getting started with or advancing in AI application development.

Packt Publishing released 'Building Production AI Agents for the Web' by Christoffer Noring, a practical guide covering LLM integration, RAG pipelines, MCP, ReAct reasoning, multi-agent patterns, and building production-ready agent harnesses.

讨论在将 AI 助手接入公司内部数据时常见的权限泄漏问题，并给出了一份检查清单和测试方法，包括按用户身份查询源数据、字段级权限、总量与排名泄漏、缓存与日志的同等保护、以及用假记录和测试用户做回归测试。

LlamaIndex 本周论文推荐介绍了 VL-DocIR 长文档检索新论文，并展示 LlamaParse Agentic Plus 在表格跨列结构解析上的准确性，可试用该工具。