Netflix describes how its knowledge‑graph search moved from rule‑based queries to an AI‑powered system that uses embeddings and reinforcement learning to personalise results. The blog outlines challenges like cold‑start content and context‑aware recommendations and explains how graph neural networks and offline/online evaluation pipelines improved discovery.
Pinterest shares how modelling user behaviour as sequences (rather than static profiles) improves the diversity and relevance of ad candidates. The post covers the transition from collaborative filtering to recurrent neural networks and sequence‑aware embeddings, with improvements in click‑through rates and latency.
Feast added native OpenLineage integration, allowing teams to automatically record end‑to‑end feature lineage across their ML pipelines. The integration uses Marquez as a backend and tracks feature definitions, materialisation jobs, and data sources, enabling unified visibility across systems and simplifying compliance and debugging
Anthropic explains why evaluating agentic systems is hard and proposes a framework that breaks evaluations into tasks, trials, graders and outcomes. The article emphasises that tests should be recorded via transcripts and outcomes, enabling reproducibility and continuous improvement
Lambda joins the Open Compute Project Advisory Board to help define standards for high‑density, composable AI infrastructure. The post outlines how Lambda’s production experience in building GPU data centers can inform reference architectures, power delivery standards and hybrid cooling frameworks
An opinionated guide on how to run large batches of LLM queries reliably. The post discusses queue‑based architectures, exponential backoff, cost control, and test harnesses to ensure outputs meet quality requirements.
An in‑depth talk exploring how RL can automate bug fixing, performance tuning and other software engineering tasks using program state as feedback.
An open‑source repository of system prompt templates for Claude models. The repo provides task‑specific instructions for code generation, testing, refactoring, and debugging, enabling consistent prompt engineering.
Moltbook introduces an open‑source platform for running agentic LLM workflows locally. It provides a notebook‑like interface, a pipeline builder and integrations with local embeddings, allowing developers to experiment with retrieval‑augmented generation and multi‑agent chains without sending data to the cloud.
MLOps Community webinar covering best practices for serving LLMs at scale, including latency, throughput and cost trade‑offs.
Databricks hosts a series of foundational workshops for practitioners beginning their data and AI journey.
A Google Cloud event discussing how organisations can leverage digital sovereignty to build compliant, resilient cloud architectures.