InfoQ

#RAG gave #LLMs access to external knowledge - but standard retrieval systems still struggle. #GraphRAG takes a different approach by using knowledge graphs to map entities, relationships, and provenance. The result? 🔹 Better retrieval quality across interconnected data 🔹 More traceable reasoning for verifiable AI outputs 🔹 Fine-grained control over what gets retrieved (and why) Watch the #InfoQ video by Cassie Shum for more insights: https://bit.ly/4vR1fVa #AI #KnowledgeGraphs #LLMs

HeroUI v3 is out. The release includes 75+ components, with 21 new additions, plus a new React Native library featuring 37 components. Built on React Aria and Tailwind CSS v4, it focuses on accessibility and customization. 🔗 Learn more: https://bit.ly/44NpHdO #React #ReactNative #CSS #WebDevelopment #SoftwareDevelopment #InfoQ

If you are running a #RAG pipeline on embeddings alone, you are leaving retrieval quality on the table. To maximize accuracy, you need to: ➤ Add BM25 ➤ Fuse with Reciprocal Rank Fusion (RRF) ➤ Consider a cross-encoder re-ranking stage 📰 Read the #InfoQ article by Aaditya Chauhan for more information: https://bit.ly/3RMIBOV #AI #GenerativeAI #LLMs #VectorDatabases

Target built a new GenAI system to overhaul marketing campaign forecasting by retrieving and ranking similar historical campaigns. Instead of relying on rule-based workflows, it uses embeddings, vector search, and LLM-based ranking to identify the most relevant past campaigns. Read #InfoQ to see how the system performed 👉 https://bit.ly/4vPljqY #AI #GenerativeAI #LLMs #SoftwareArchitecture #RAG

A restaurant deployed a chatbot. When a user asked it how to solve a programming problem, the system answered. That example captures the usage problem. Once real users are in the system, they'll enter inputs it was never designed to handle, and you can't reproduce that in a sandbox. Latency and cost are the second issue, and neither surfaces until production load hits. By then, the architectural decisions that caused them, made months earlier during prototyping, are expensive to unwind. Most teams building AI systems right now are working through both for the first time, without peers outside their own company to pressure-test decisions before they get locked in. The InfoQ Certified AI Engineering Program is five weeks online with a small cohort of senior engineers and software architects, facilitated by Hien Luu, author of MLOps with Ray and QCon Program Committee Chair. First cohort: July 25, 2026. Places are limited. Link in the comments. #AIEngineering #ProductionAI #LLMOps #SoftwareArchitecture

The gap between AI that works in a demo and AI that works in production shows up in five specific places. Week 1 sets the baseline on where AI adds value in your SDLC and where to deliberately keep it out. Week 2 is retrieval design and knowledge graphs, covering how to build RAG pipelines that stay grounded rather than confidently wrong. Week 3 covers agent design, from single-purpose tools to multi-agent orchestration, with the trade-offs between autonomy and certainty built into the design from the start. Week 4 covers inference gateways and GPU cost controls, and the question of what to centralize versus keep federated. Week 5 is evals, trust, and rollout readiness, with a group capstone presentation that ends on an unresolved architectural question your cohort is still working out. Four hours a week, five weeks, facilitated by Hien Luu, author of MLOps with Ray. Next cohort starts July 25. Saturdays 09:00 AM PDT. Link in the comments. #AIEngineering #MLOps #InfoQ

#AgenticAI architecture is redefining how software is designed - and it's poised to shape the industry for years to come. This #InfoQ eMag explores the core concepts, patterns, and emerging trends driving agentic AI as it moves into the mainstream. 📥 Download your free copy here: https://bit.ly/4giAG6i 👏 Big thank you to all our brilliant contributors: Mallika Rao, Karthik Ramgopal, Adi Polak, Subash Natarajan, Ahilan Ponnusamy, and InfoQ editor Rafał Gancarz. #SoftwareArchitecture #AIArchitecture #AIAgents

Logs remain one of the richest sources of operational insight, but as AI workloads drive telemetry volumes higher, turning that data into actionable intelligence is becoming increasingly difficult. Join InfoQ and Datadog on July 9 for a technical session introducing a practical maturity model for AI-assisted observability. You'll learn: • Why centralized log pipelines are the foundation for effective observability • How AI-powered anomaly detection is changing incident response • What natural language log investigation can realistically accomplish today • Strategies for controlling telemetry costs without sacrificing signal quality The session includes real-world examples and live AI-assisted investigation workflows that your team can apply regardless of your observability stack. Register for free: https://lnkd.in/dr3AjHK8 #Observability #SoftwareEngineering #PlatformEngineering #SRE #DevOps #AI #CloudNative Sponsored by Datadog

Software architecture is as much about people and business as it is about technology. In this #InfoQ podcast, Sonya Natanzon breaks down the critical intersection of technical and social architecture. Key takeaways: • Business > Tech: How the company operates matters • Focus on Outcomes: Good requirements look at problems to solve • Embrace Constraints: Boundaries eliminate decision fatigue 🎧 Listen now: https://bit.ly/4fp59iv 📄 #transcript included #SoftwareArchitecture #RequirementsEngineering #DDD #TechLeadership

Elastic has open-sourced Atlas, a memory system for AI agents built on Elasticsearch. Key features: • 3 categories of memory • MCP integration • Per-user memory isolation How does it perform? Atlas hit a 0.89 Recall at 10 on question-answering tasks. 🔗 Read more: https://bit.ly/4veT6sl #AI #Elasticsearch #AIAgents #MCP #OpenSource #InfoQ