MLOps news and analysis on Web Pulse: 3359 curated articles tracking the latest MLOps developments, tools, and research, updated continuously from vetted sources.
A developer's technical breakdown of local LLM memory requirements shows that a 70-billion-parameter model needs roughly 140 GB of VRAM at FP16, 70 GB at INT8, and about 35 GB at 4-bit quantization, meaning even an 80 GB…
A developer maintaining RetroPrompt, an AI portrait image-generation project, described a quota-handling design built on a Cloudflare Worker and D1 that distinguishes confirmed failures from unconfirmed outcomes by track…
A developer reworked the cost math for Aria, a Go-based text-to-SQL CRM assistant that uses RAG over the database schema, after Anthropic cut Claude Opus 5.5 cache reads to $0.20 per million tokens, 60% below Opus 5. The…
A developer argues that conventional RAG architectures create a dangerous chasm between the data lakehouse and the AI stack, forcing teams to duplicate data and reimplement security in a separate vector database. The pos…
In the sixth essay of his AI Harness Engineering series, developer Derek Wang argues that resilient implementation — idempotency, graceful degradation, visibility, and fast recovery — is the foundational layer of any AI-…
Developers at Mininglamp built Mano-AFK, an open-source autonomous execution framework that takes a natural language prompt through PRD generation, coding, local deployment, three-layer testing, and automatic bug fixing,…
A new arXiv paper (2609.25237v1) proposes a post-training delivery benchmark that tests whether LLM agents can be trusted as forward-deployed engineers (FDEs), rather than whether they can raise a metric. The benchmark's…
Bengaluru lab Conscious Engines released Kinetic-4B, a 4-billion-parameter model that scored 82.33% accuracy at 1.61s p95 latency on a 300-sample Composio tool-calling benchmark, edging out Anthropic's Claude Haiku 4.5 (…
A technical explainer outlines how data engineering underpins modern AI and data science, describing it as the software discipline responsible for collecting, storing, and processing data at scale. The piece traces the f…
Mubit launched as an agent decision attribution and calibration platform that tracks and adjusts AI agent decisions per user at scale, according to its founder's Show HN post. The platform lets operators view an agent's …
Croft has added built-in monitoring and telemetry to every app in its workspace, requiring no SDKs or code changes. The company says logs, errors, and telemetry are automatically surfaced to a user's AI so it can diagnos…
A quantization report on MiniCPM5-2B found that the Q6_K GGUF quantization is indistinguishable from the F16 baseline, while IQ4_XS is the last usable quant before a sharp quality cliff into the Q3 region. The testing, p…
NVIDIA built a five-stage "Task-Seeded SDG" pipeline for its Nemotron model family, drawing roughly 70 public tasks and about 700 subtasks from lm-eval-harness to generate synthetic QA pairs that match seed tasks' skill …
A developer built a demo showing how to stop runaway AI agents from overspending by attaching owner and spend-limit claims to machine-to-machine JWTs issued by Kinde and enforcing them against a ledger in Convex. The bui…
Linear reworked its CI pipeline after AI coding agents made it a bottleneck, roughly quadrupling its test suite while cutting PR wait time from over six minutes to just over five and halving runner time per test. The cha…
Two engineering teams reported that AI coding agents shifted their bottleneck from code generation to code validation. Linear reworked its CI pipeline after test suites nearly quadrupled, cutting pull request wait time f…
A latent-GRPO training step for the Qwen 3.8 27B model dropped from a median of 36 minutes on a single H100 to 2.6 minutes on one 8× NVIDIA B300 node, a 14× speedup, according to the training team's account of a dozen fi…
An audit of 110 open-source AI usage-metering tools uncovered more than 45 verified bugs across five recurring families, including stale pricing tables, provider-wrong cache multipliers, and retry double-counting in stre…
Amazon Web Services launched Amazon CloudWatch Omni, an AI-powered observability experience that organizes monitoring around applications rather than individual signals and is built on OpenTelemetry. CloudWatch Omni prov…
Linear's CTO detailed how agentic coding drove the company's test suite to nearly quadruple, with roughly 2,000 new tests added weekly, forcing months of pipeline performance work to keep pull request wait times from gro…