RAG
30 articles tagged with this topic
AI Engineers' Real Barrier Isn't LangChain—This Project Lays Bare the Stack
calmrocks' zero-framework Colab tutorials went viral on GitHub. We're watching the deeper signal: the AI engineer role is stratifying by who truly und
Testers Become AI Quality Inspectors: Open-Source Roadmap Exposes a Talent Gap
A 13-chapter AI testing roadmap hit GitHub, gaining thousands of stars. Real signal: "who verifies AI" is becoming a new job.
AI's Bottleneck Has Moved: From Model Selection to the Real-Time Data Pipeline
Model selection isn't AI's bottleneck. We see the real-time data pipeline — most enterprises stuck on T+1 batch — blocking Agents in production.
RAG now needs monitoring — SMB AI apps shift from 'just works' to 'can't crash'
A Juejin post shows lightweight RAG monitoring: 5s probes, 3s timeouts, WeChat Work alerts. AI apps shift from 'can it work' to 'must not crash'.
90% of AI Agents Are Wrappers — Three Tests to Filter Vendor Hype
Real agents have clear technical definitions. Three hard metrics from frontline practice to help buyers cut through vendor pitch in AI Agent selection
Doubao's Zhiya: Teaching AI 'I Don't Know' Is Harder Than Making It Right
Zhiya, a Doubao-ecosystem workbench, silences AI when sources can't be verified—turning trust from a model problem into an engineering one.
Corporate AI Exam Generation: PPT to Paper in Hours, But Trust Gap Remains
How LLMs+RAG fix enterprise exam creation's three pain points—slow, person-dependent, outdated—representing AI's first real unglamorous production win
AWS Turns Enterprise Knowledge Management Into an AI Template
AWS releases an RAG-based AI template for capturing veteran expertise — a key standardization signal for manufacturing, healthcare, and energy.
AWS Bedrock RAG Fix: Pre-Filter Docs With Cheap Model—Save Money, Buy Complexity
RAG queries feed Claude Sonnet 5-20 chunks per ask. AWS proposes pre-filtering with cheap Claude Haiku. Saves tokens, adds complexity.
Mistral's Agentic Search: 86% accuracy on complex docs—at 71-second cost
Mistral's Agentic Search hits 86% accuracy on FinanceBench, but each query now averages 71s—shifting from snippets to evidence chains.
Beyond Tool-Calling: The Real AI Agent Divide Is Decision Memory
Tool-calling is just the entry ticket. Real business agents must write every decision back to form auditable, learnable memory — the 2026 enterprise A
The Hidden Cost of Enterprise AI Knowledge Bases: Every Update Re-burns Compute
Enterprise AI knowledge bases reindex on every document update. Million-doc rebuilds take hours; incremental updates risk conflicts. The RAG pitfall.
GraphRAG & HippoRAG Tested in China: Promising Theory, Painful Deployment
GraphRAG and HippoRAG promise AI multi-hop reasoning via knowledge graphs. A Chinese test on 89 enterprise QA questions reveals: promising theory, pai
Agent Model Selection: Break Down Tasks First, Then Compare Parameters
A Java developer forced into Agent work asked "which model is strongest" — exactly the wrong question. Four model types, not one, run a real Agent sys
Ragent 1.1.0 Ships: A Chinese Developer's RAG Framework Goes Enterprise-Grade
Ragent 1.1.0 ships from a solo Chinese dev. What was 'runnable' becomes 'auditable': four-channel hybrid retrieval, source attribution, tiered routing
AI Agent Tool Selection: Don't Dump Everything In — An AI Pharma Paper's Fix
Three tool-filtering modes for AI Agents — LLM, embedding, all-in. An AI pharma paper exposes real cost-vs-accuracy tradeoffs in enterprise rollouts.
Tencent Open-Sources EVIE: Storage Cut to 1/8 — China Enters RAG Race Head-On
Tencent's open-source EVIE document retrieval cuts per-page storage to 1/8 with higher accuracy — Chinese AI labs now competing globally in enterprise
DeepSeek Cache Prices Surge 12x — Companies Squeezing AI Costs Must Recalculate
DeepSeek's new pricing (Aug 17): V4-Pro peak cache hits jumped from ¥0.025 to ¥0.30 per million tokens — 12x. A targeted strike on high-frequency RAG
Agent Learning's Real Threshold: Reciting Concepts ≠ Using Them Correctly
After Agent basics, a developer's 6 Q&As: Chatbot vs Agent, streaming Tool Calling, Loop stopping, context trade-offs. Demo proves chain, not producti
Vector Database Selection: The Underestimated Bottleneck in Enterprise AI
We read a vendor selection guide: most enterprises built their RAG stack on the wrong vector database — that's why AI answers off-topic despite powerf
DeepSeek Slashes Cache Pricing to 1/10 — LLM Application Layer Reshaped
DeepSeek's cache-hit pricing drops below one cent per million tokens — 1/10 of mainstream models. The first Chinese LLM vendor to reshape Agent and RA
Why Enterprise AI Gets It Wrong: It's Not the Model, It's Document Splitting
Deep-dive on RAG's most overlooked step: document splitting. When AI misreads your docs, the chunker — not the LLM — is usually to blame.
Enterprise AI Knowledge Bases Miss the Mark: 80% of Work Is Document Loading
Enterprise RAG projects fail because documents aren't AI-ready. LangChain's Document abstraction solves this hidden 'first mile.'
Graphify Uses Knowledge Graphs to Make AI Read Code Instead of Guess
Y Combinator-backed Graphify hits 73k GitHub Stars in 2.5 months by replacing vector search with code knowledge graphs for Claude Code and Cursor.
Notion Open-Sourced the Strongest Reranker — But That's Not the Real Story
Notion open-sourced Zerank 2 after acquiring Zeroentropy, matching paid reranker APIs. For builders, this drops the cost of local RAG pipelines anothe
Build a Local AI Knowledge Base: LangChain + Ollama Make PDF Q&A Simple
A hands-on guide using LangChain with Qwen2 and bge-m3 to build an offline RAG knowledge base that answers PDF questions on your own machine.
Two AI Open-Source Projects Tested for Cross-Repo Analysis: Result Was Zero — And That's the Right
LightRAG and graphrag are peer tools in graph-based RAG, neither calling the other. A zero-result cross-repo test reveals a real distinction.
Milvus 2.6 Completes the Storage-Compute Split — Vector Databases Get Serious About Engineering
Milvus 2.6 decouples compute from object storage, etcd, and message queues — a sign AI infra is maturing beyond Embedding model benchmarks.
The Hidden Cost of Code Knowledge Bases: Incremental Updates Are Ten Times Harder Than Full Rebuilds
Building a code-searchable AI knowledge base is easy; keeping it accurate three months later is the real challenge. Incremental update strategy is bec
Enterprise RAG in Practice: The Bottleneck Is Docs and Retrieval, Not Models
A real enterprise deployment shows RAG succeeds or fails less on model size than on document parsing, chunking, retrieval, and citations.