InstacartFrom print to digital: Making weekly flyers shoppable at Instacart through computer vision and LLMs
This post details a two-phase AI pipeline that digitizes grocery flyers into interactive, shoppable experiences on Instacart by merging custom image segmentation (SAM-based) with OCR and LLM-powered product identification, enabling fast, scalable flyer digitization.
AWS MLAutomated Reasoning checks rewriting chatbot reference implementation
An open-source chatbot reference implementation that uses Automated Reasoning checks to iteratively rewrite answers, verify correctness with mathematical proofs, and surface auditable explanations through a Guardrails-enabled pipeline.
OpenAITesting ads in ChatGPT
Technical guide on evaluating ad integrations within ChatGPT, covering testing approaches, metrics, and best practices.
SalesforceHow Agentic Memory Enables Durable, Reliable AI Agents Across Millions of Enterprise Users
Explores how Agentic Memory within Agentforce externalizes durable, governable memory into a structured enterprise memory platform - enabling adaptive context, episodic reasoning, and hybrid validation to keep AI agents reliable across millions of users.
Google CloudOTLP everywhere: Cloud Monitoring now supports OpenTelemetry Protocol metrics
Cloud Monitoring now supports OTLP-based metrics via OpenTelemetry, enabling a provider-agnostic telemetry pipeline, advanced features like Delta metrics and exponential histograms, direct SDK-to-Cloud Monitoring ingestion, and a managed OpenTelemetry option for GKE with zero-code instrumentation.
SpotifyHow We Release the Spotify App: A Look Under the Hood (Part 2)
An in-depth look at Spotify's Release Manager Dashboard (a Backstage plugin) that unifies release data via an API gateway, automates state transitions with a Robot, and enables multi-platform releases from development to production.
MetaBuilding Prometheus: How Backend Aggregation Enables Gigawatt-Scale AI Clusters
Designing and deploying Backend Aggregation (BAG) to stitch thousands of GPUs across regional data centers into Prometheus, a gigawatt-scale AI cluster, using Disaggregated Schedule Fabric and Non-Scheduled Fabric L2 networks for high-capacity, resilient interconnect.
Google CloudUNC1069 Targets Cryptocurrency Sector with New Tooling and AI-Enabled Social Engineering
Technical analysis of UNC1069's macOS-centric intrusion into the cryptocurrency sector, detailing a new tooling suite (WAVESHAPER, HYPERCALL, HIDDENCALL, DEEPBREATH, CHROMEPUSH, SILENCELIFT, SUGARLOADER) and an AI-enabled social-engineering lure (compromised Telegram, a fake Zoom meeting) used to harvest credentials and browser data for financial theft and future campaigns.
How Duolingo teaches speaking skills
A technical overview of Duolingo's approach to teaching speaking skills through integrated speaking exercises, mic-based practice, and real-time conversation tools.
AWS MLAgent-to-agent collaboration: Using Amazon Nova 2 Lite and Amazon Nova Act for multi-agent systems
Orchestrating travel planning with a three-agent architecture that uses Amazon Nova 2 Lite for reasoning and planning and Amazon Nova Act for browser automation to transform a fragile single-agent solution into a robust, explainable multi-agent system via simple A2A message-passing.
AWS MLAccelerate agentic application development with a full-stack starter template for Amazon Bedrock AgentCore
FAST provides a ready-to-deploy full-stack starter template for Amazon Bedrock AgentCore that accelerates agentic application development by wiring runtime, memory, tools, authentication, and observability into a production-ready stack with infrastructure-as-code via AWS CDK and a React frontend.
AWS MLNew Relic transforms productivity with generative AI on AWS
New Relic builds NOVA on AWS to transform internal productivity with generative AI, using Bedrock-backed RAG, Kendra, and MCP-driven Strands Agents to automate complex workflows across Confluence, GitHub, Slack, and Jira while maintaining security, sub-20-second latency, and measurable productivity gains.