MLOps news and analysis on Web Pulse: 3308 curated articles tracking the latest MLOps developments, tools, and research, updated continuously from vetted sources.
A developer built Origin, a repository interface that tracks AI-written code by agent, model, prompt, cost and review status, reporting that 93.9% of code in their own repo was written by Claude. Origin also stores per-r…
A developer outlines a .NET edge-to-cloud time-series architecture for water and power meter fleets, using TimescaleDB hypertables, Hypercore columnstore compression, and continuous aggregates to keep readings tied to lo…
The MonkeysCode team shipped two infrastructure updates to its context kernel and model proxy to harden autonomous coding agents against dropped SSE streams and prompt-cache misses. The proxy now injects SSE comment ping…
Gremlin announced the general availability of Gremlin Foresight AI, a product the company says identifies reliability risks before they become incidents and verifies the fixes, following a successful beta. Gremlin CEO Ko…
A distributed multi-agent setup running llama-server instances across dedicated LAN machines needs a separate communication and orchestration layer rather than relying on llama.cpp for coordination, according to advice g…
A developer argues that storing derived AI cost metrics such as cost per successful outcome is insufficient to explain why those metrics change, since a drop from $1.80 to $1.20 could stem from lower attributable cost, m…
A developer built @sbissoli/mcp-surface, an npm tool that locks an MCP server's declared and no-token surfaces into a surface.lock.json file and fails tests, deploys or publishes when the surface changes without a versio…
The Open Source Technology Improvement Fund published the results of a security audit of vLLM conducted by X41-DSec, which found 5 security-impact findings — 1 critical and 4 high — plus 15 findings without direct securi…
Instacart's machine learning engineering team published a case study on an agentic modeling loop in which autonomous agents formulate hypotheses, write training code, launch distributed training jobs, and evaluate result…
Honeycomb published a complete, annotated OpenTelemetry Collector configuration for LLM observability that receives traces over OTLP, normalizes OpenInference and OpenLLMetry schemas onto GenAI semantic conventions, reda…
A developer's technical breakdown explains why full-parameter fine-tuning of an 8-billion-parameter model in 16-bit precision exceeds 80 GB of VRAM, since AdamW training consumes roughly 16 bytes per trainable parameter …
A weekly LLM inference digest highlights new research aimed at KV cache and MoE serving bottlenecks, including TaSQ, a 1-bit KV cache quantization method that its SGLang implementation on a single RTX 6000 Ada shows supp…
Miren released ephemeral deployments, also called pull request environments, that let CI agents such as Claude Code review pull requests against a full application deployment instead of a constrained CI environment. The …
Vercel AI Gateway has added confidence-based decision fallbacks to its experimental_decide feature, letting applications route a successful but low-confidence decision to a second model instead of accepting it or escalat…
PagerDuty announced the general availability of the Runbook Automation / Rundeck MCP Server for Rundeck and self-hosted Runbook Automation customers on version 6.1.0 or higher, letting AI assistants in Claude, Cursor, VS…
A new recipe for running DeepSeek V4.1 Flash on a single AMD Strix Halo machine (gfx1151 GPU, 128 GB unified memory, several NVMe drives) reports roughly 450 tokens per second prefill and 15 tokens per second decode with…
A hackathon organizer outlines an evaluation protocol for testing AI prototypes before final pitches, using a 30-announcement illustrative dataset split into 18 team-visible and 12 held-out examples for a JSON field-extr…
A developer built NovaFabric, an observability system that captures AI agent executions into portable, tamper-evident "Run Capsules" containing traces, model calls, tool calls, environment locks, and replay metadata. The…
A developer published a Python statistics checklist for evaluating claims that an LLM endpoint has been degraded, using Wilson confidence intervals, Fisher's exact test, and a 200,000-trial simulation showing that with 1…
Anyscale on Azure became generally available on October 7, 2026, letting enterprises run the full AI loop — data processing, training, post-training, evaluation, and inference — on Azure Kubernetes Service in their own s…