AI & Machine Learning

Model training and serving, retrieval pipelines, and the infrastructure underneath them — LLMs, RAG, embeddings and vector search, alongside longer-standing ground like computer vision and NLP. Most of it is teams describing systems already carrying real traffic: what they built, what it cost, and where it broke first.

Loading
Building the Agent Host Protocol for sessions that continue across devices
GitHub describes the Agent Host Protocol (AHP), an open standard that keeps Copilot agent sessions continuous across VS Code, the GitHub Copilot app, mobile apps, and cloud sandboxes. The post explains why per-client integrations failed, how a URI-channel, reducer-based state design handles reconnection and concurrent mutations, and what was learned from building two independent hosts in TypeScript and Rust. It also covers how the protocol stays agnostic to agent behavior and where to connect...
Smooth Cruiser at Lyft: Built for Safety and Transparency
Lyft describes how Smooth Cruiser, its driver safety scoring feature, measures driving signals relative to local road conditions, regional peer groups, and a rolling window of recent rides. The post also covers the data platform behind the score, including ML-based harsh event detection, an Apache Airflow ETL pipeline, and DynamoDB for low-latency score retrieval, along with transparency features that show drivers the rides behind their scores.
DuckDB User Group Workshop on AI-Powered Review Analysis with Hugging Face
A DuckDB User Group workshop built a pipeline over 2.1 million Amazon appliance reviews to find products with falling ratings. DuckDB SQL selects the largest rating drops and the most helpful recent complaints, and only those 40 reviews go to an LLM for theme summarization. The post also covers running the job on Hugging Face and deploying the workflow as a Gradio app with API endpoints.
Introducing eve 1.0
Vercel announces eve 1.0, an open-source framework for production AI agents, and details the engineering behind it. Median hosted turn latency fell from 3.04s to 414ms through a single-orchestrator workflow design and WebSocket event writes, and estimated workflow cost for a 50-turn session dropped about 76% by storing only changed conversation data. The post also covers OpenTelemetry-based tracing, multi-tenant agent configuration, and dynamic tool loading.
Using a "chaos loop" to harden the Claude Docs launch: 60 server errors across billions of requests
Vivek Nair describes how a team hardened the Claude Docs backend before its September 2026 launch by running Claude in a long-lived loop against a chaos-testing staging cluster. Key lessons include using copy-on-write to fix per-keystroke memory growth, which cut projected pods from over 500 to about 50, and choosing Rust with minimal dependencies so telemetry traces back to owned code. The launch saw 60 server errors across 2 billion requests with no rollbacks.
Engineering AI Discover
Netflix describes the intelligence layer behind its AI Discover TV feature, an agent harness that combines LLMs with recommendation systems. The post covers a plan-and-execute architecture with controller and executor components, validated plans built from composable capabilities, model-agnostic design, asynchronous page generation, and safety rules enforced in code rather than prompts. It is an engineering write-up of the trade-offs and decisions made along the way.
How AI Slashed 600-Test-Failure Triage From Days to 90 Minutes
Salesforce engineers describe building AI-Assisted Test Failure Buddy, a Python and FastMCP tool exposing eight MCP tools to Claude Code and Cursor, to triage over 600 test failures caused by retiring the SOAP API login mechanism. The post explains how deterministic code handles API calls, batching, signature matching, and reassignment while AI interprets unmatched stack traces and ambiguous ownership, reducing estimated two to three days of triage to about 90 minutes.
Split-and-Scale Latent 3D Representations
Presents split-and-scale, a method that increases latent capacity of pretrained 3D tokenizers at inference by applying the same tokenizer to smaller sub-regions of a scene, since 3D scenes are represented as continuous functions of coordinates. The authors ablate compute, memory, latent capacity, and reconstruction quality, and report competitive results against larger models exceeding single-GPU memory, plus an autoregressive 3D generator built on the approach.
From Predictive to Generative – How Michelangelo Accelerates Uber’s AI Journey
Uber's engineer write-up traces the evolution of its centralized ML platform, Michelangelo, across three phases from 2016 to the present. It details the shift from tree models to deep learning, including the move from Spark to Ray-based trainers, elastic Horovod, and Triton model serving, along with the platform changes that followed. It also covers model quality tiering, Canvas for code-driven pipelines, and the Gen AI Gateway for LLM access.
Tabular foundation models on Databricks AI Runtime
The post explains how tabular foundation models (TFMs) use in-context learning and row-compression architectures such as TabPFNv2 and TabICLv1, and compares their licensing and GPU inference costs. It then describes how Databricks AI Runtime addresses GPU provisioning, observability, and data-governance challenges for TFM experimentation and fine-tuning.
Give sandboxed agents live web access with the Firecrawl kit for Docker Sandboxes
The post introduces a Firecrawl kit for Docker Sandboxes that gives isolated coding agents live web search, scraping, and crawling through one allowlisted host, api.firecrawl.dev. It explains how the kit's mixin descriptor declares install, credential, phased network-policy, and agent-context capabilities, with the API key injected by a host-side proxy so it never enters the sandbox. It also covers verification steps and known limitations such as proxy 403s and organization policy overrides.
How the World’s Most Advanced AI Infrastructure Supply Chain Invents the Future
NVIDIA's operations leader Deb Shoquist describes how the company builds supplier relationships to scale AI infrastructure production. The post walks through a GB200 compute board test-handler throughput failure, the cross-team response that followed, and a traceability process that cut chip-to-system lead time from five weeks to two. It is written by NVIDIA about its own manufacturing operations.
Maintainers don’t need more AI bug reports. They need fixes.
GitHub and Microsoft ran a pilot using Codename MDASH, an AI scanning harness, to find vulnerabilities in six open source projects. An AI triage agent removed duplicates and researchers verified each finding by hand before reports went out privately with proofs of concept and suggested fixes. The post distills lessons for anyone sending AI-generated security reports: verify findings, report privately, and include a fix.
Agents are users, so your software needs evals
The post argues that software used by AI agents needs evals, meaning repeated, graded runs rather than single tests, and lays out an agentic testing pyramid. It walks through building an end-to-end eval with Harbor on Vercel Sandbox and AI Gateway, including choosing between pass@k and pass^k, designing tasks and verifiers, and analyzing traces to find failure causes.
How the Claude Code team thinks about verification
The post explains how the Claude Code team uses end-to-end, property, and differential tests to let coding agents verify their work with evidence. It also describes the practices used to keep verifiers honest, including a commit gate, a reliability checklist tied to diffs, and evaluations that measure review recall against past incidents.
Introducing the Databricks AI Index: The state of enterprise AI tokenomics
Databricks introduces the AI Index, a public resource combining anonymized enterprise usage data from Unity Gateway with coding-agent benchmarks. The post reports that open-weight model traffic rose from 7.3% to 30.9% and that cost per completed task can diverge sharply from token price, with one model costing about 32x more than another at equal success rates. It also describes the benchmark methodology, including hidden tests and publicly listed prices.
A step toward defining the Standard Model on a lattice
The post body was unavailable, so this description is based only on the title and feed summary. It describes a candidate nonperturbative lattice definition of the Standard Model and says GPT-6 Astra helped develop the idea into a detailed proof. Treat the content class and education score as provisional until the full text can be checked.
Your coding agent shouldn't grade its own homework
Momentic's engineering team argues that coding agents should not verify their own work, citing an audit where only 9.6% of 1,000 AI-authored unit tests were useful and a comparison showing agents that wrote tests did no better than those that did not. They describe an independent QA agent that tests live applications from the original requirements, which their benchmark found surfaced 3-5 times as many bugs as agents checking their own output.
Advancing science & engineering with Antigravity’s /teamwork
Google Research describes results from Antigravity's /teamwork multiagent system across mathematics, chemistry, systems software, and ML engineering. The post explains design choices such as partitioned context, dedicated falsifiers like the Lean 4 checker, and verified outcomes including a Courtade–Kumar conjecture proof and a Rust port of libgav1. It also notes which tools are available to external researchers and developers.
Build a real-time next-best-offer recommender using Databricks Feature Views
The post walks through a Databricks Solution Accelerator that uses Feature Views to define batch and Kafka streaming features once for both training and serving, reducing train/serve skew. It reports benchmark results for a real-time next-best-offer recommender, including latency, feature freshness, and throughput, along with a concurrency sizing formula and the measurement limitations.