LLMs hallucinate in domain-specific languages (DSLs). Typed Domain Grounding offers a way to reduce it. Irakli Betchvaia explores how embedding DSLs in mainstream typed languages, paired with compiler…
無論是哪種資訊系統、不管是 architecture 還是 infrastructure 都歡迎討論
@information-system-design@baudrate.tw
Search for this handle on your own server to follow from there.
LLMs hallucinate in domain-specific languages (DSLs). Typed Domain Grounding offers a way to reduce it. Irakli Betchvaia explores how embedding DSLs in mainstream typed languages, paired with compiler…
Why do almost all Python JIT compilers die? Building a JIT is massive: ⚡ HotSpot: 1,000s of person-years ⚡ V8: ~50 full-time engineers Laurence Tratt explains why dynamic languages make JITs a sinkhol…
How do you find the root cause of a large-scale cloud incident? Atlassian’s new approach correlates metrics, logs, distributed traces, and service topology to rank hypotheses about where failures star…
Agoda moved its 1.5 TB hotel Price Cache from 72 SQL Server shards to DragonflyDB to handle rising read/write volumes. The result? ~8× lower P99 read latency, with two DragonflyDB clusters providing h…
It’s been a busy week in the #Java ecosystem! Highlights include: ➤ New JEPs for ahead-of-time compilation and structured concurrency; GA releases of Jakarta CDI 5.0 and ADK for Kotlin 1.0; the Septem…
Think an EU cloud region protects your data from US jurisdiction? Physical server location ≠ data sovereignty. Martin Kleppmann explains why US law still applies to European data hosted by US cloud pr…
The ESP32 Bit Pirate project uses HTML5 APIs to install firmware and interact with ESP32-S3 boards, supporting multiple digital and wireless protocols. 🛠️ Includes hands-on guides for memory dumping, s…
Think long context windows & vector search make #AIagents production-ready? Think again. Cassie Shum explains how #KnowledgeGraphs can become the operational substrate for agentic systems - providing …
#Postgres can be the durable state store and coordination layer for workflows, eliminating the need for an external orchestrator. 🔹 SKIP LOCKED enables concurrent work 🔹 Primary-key checkpoints enforc…
A building collapses when one structural failure triggers a domino effect. Software systems can fail the exact same way - except our dominoes are hidden. Sam Newman explains why microservices, third-p…
~700 OpenAI agents in a cybersecurity benchmark found a way to communicate via a secret message board - exchanging 70,000+ messages in 6 days. Instead of working independently, they coordinated exploi…
As you get more senior, architecture calls can become harder to test outside your own organization. The InfoQ Software Architecture cohort gives you a senior peer group to do exactly that: https://cer…
#Cloudflare’s Cache Transcoding prototype uses Zstandard to compress cache content - HTML, JSON, CSS, JavaScript - before storing it on disk. The potential payoff: petabytes of additional effective ca…
AI is taking over the routine tasks junior devs used to learn from. Scott Hanselman proposes a healthcare-inspired preceptorship model: 🔹 Senior devs focus on mentorship 🔹 Success = developing talent,…
Consistency is the floor. Accountability goes further. Building AI for high-stakes, regulated environments? Prompt logs & model versions aren’t enough. True auditability requires explaining why an inp…
Meta recently announced the beta release of Astryx - an #opensource React design system developed internally. Built on React 19 + StyleX, Astryx includes: 🔹 150+ accessible UI components 🔹 Customizabl…
tsgolint v7 is here, bringing Go-powered, type-aware linting to Oxlint. ➡️ 12–18x faster than ESLint + typescript-eslint ➡️ 59 of 61 type-aware rules supported ➡️ Native TypeScript 7 integration via type…
🚨 90% of enterprise AI agents are stuck in POC! The bottleneck? Teams don't know when an agent is reliable enough for production. In this #InfoQ talk, Zhou Yu shares: 🔹 Why agents stall in the demo ph…
#Netflix reworked its Conductor workflow engine to handle workflows with up to 30,000 tasks (up from ~2,500) while cutting p99 evaluation latency by ~40%! Key updates in Conductor 4.0: 🔹 Separates wor…
The real problem with enterprise data? Not multiple definitions - pretending there’s only one. If Marketing and Finance can’t agree on what an “active customer” is, how can an AI agent? Enterprise AI …