MLOps news and analysis on Web Pulse: 3347 curated articles tracking the latest MLOps developments, tools, and research, updated continuously from vetted sources.
Llama.cpp's default layer-split mode gives a dual-GPU setup the combined VRAM of both cards but runs single-stream generation at roughly the speed of one card, while tensor parallel can nearly double single-stream throug…
A developer built SHIPCHECK, an autonomous ReAct agent that acts as a pre-flight production release gatekeeper by querying a live Sanity Content Lake through a typed MCP JSON-RPC 2.0 interface. The agent runs a 13-step r…
A developer compared twelve open-source and self-hosted Deep Research systems, grouping their architectures into five categories: recursive research trees, planner-plus-subagent designs, evidence-gap loops, perspective-d…
A developer benchmarked Google's Gemma 4 E2B quantization-aware trained (QAT) checkpoint against the full bf16 release on a single NVIDIA L4 SageMaker endpoint, finding the 4-bit QAT model decodes at 105.1 tok/s versus 5…
A security-focused analysis argues that the key difference between AI agents and traditional automation is where control over execution lives: in code engineers wrote, or in a model-driven decision loop. Citing Anthropic…
A developer published a step-by-step guide and open-source repository for deploying Google's Gemma 4 E2B model to an Amazon SageMaker real-time endpoint on a single NVIDIA L4 GPU using AWS's vLLM container. The project w…
A developer built a ~700-line RAG CLI in Bun and TypeScript using SQLite with sqlite-vec, Voyage embeddings and Claude generation, then found that the evaluation harness itself was broken. Switching from exact-substring …
GitHub added a `pull_request_review_times` array to its enterprise and organization repository-level Copilot usage metrics reports, breaking pull request review time into three stages: ready-for-review to first review, f…
A developer published a tutorial showing how to build a Kotlin/Jetpack Compose diagnostics dashboard for robotics and Physical AI systems, using StateFlow and a ViewModel/Repository architecture to stream battery, temper…
A developer published a tutorial showing how to build a Kotlin/Android operator layer for robotics, XR, and edge-AI systems that pairs vision-language-action models with bounded, lifecycle-aware state management. The arc…
Dataiku shipped Agent Management on September 24, a standalone, platform-agnostic product that discovers, tracks and risk-ranks AI agents across a customer's stack, with general availability set for October 2026. The lau…
Laminar is an open-source observability platform for AI agents, Apache 2.0 licensed at github.com/lmnr-ai/lmnr, that traces LLM calls, tool calls, sub-agents, tokens and cost and uses its Signals feature to cluster unexp…
Revefi identified roughly $300,000 in annual savings on a single internal AI agent for one customer after fixing prompt-caching mistakes that had sent its LLM data spend spiraling, the agentic AI platform said. Revefi's …
AWS published a reference implementation that wires Amazon Bedrock AgentCore Evaluations, generally available since March 2026, into GitHub Actions so pull requests touching agent code, system prompts, model selections, …
Collibra BV launched four runtime governance capabilities — Live Map, Maestro, Guardian Agents and Agent Contracts — to govern enterprise AI agents that take action across systems, CEO and co-founder Felix Van de Maele s…
A small open-source lab in Switzerland that runs AI coding agents across roughly a dozen projects documented five cases where agents confidently reported success while being wrong, each traced to a measurement flaw rathe…
In a webinar titled "Same Job, New Rules," Matthias Bohlen argued that AI does not remove the architect's job but adds new risks the design must control, since system behavior now comes from data, models, prompts, search…
A developer built RelayLab, a local multi-agent system that pairs a Builder model (Qwen2.5-Coder) with a Reviewer model (Qwen3) via Ollama and a Python orchestrator to autonomously develop an app over six rounds. After a…
LangChain launched LangSmith Fine-Tuning and its `smithtune` CLI, which turns agent trajectories stored in LangSmith into datasets for supervised fine-tuning (SFT) of smaller models, handling dataset creation, training v…
AWS published an architecture that combines Amazon Elastic Kubernetes Service (Amazon EKS), Elastic Fabric Adapter (EFA), and DeepEP to scale Mixture-of-Experts reinforcement learning, reporting 40% more throughput. The …