engblogs

Leverages a top-down, retrieval-augmented generation pipeline with LLMs to deliver cohesive, personalized Shopping Hub recommendations at scale, spanning design, evaluation, and ranking.

A practical blueprint for COBOL modernization on the mainframe that combines AI-assisted reverse and forward engineering with deterministic, platform-aware analysis to deliver traceable and production-ready results at scale across large enterprise portfolios.

RAG-powered Dash retrieves enterprise context and grounds answers, while a hybrid human-in-the-loop and LLM-assisted labeling approach scales relevance judgments to train its search ranking models for improved enterprise search relevance.

Serving Iceberg lakehouses data fast and fresh at scale with Spanner's columnar engine to deliver real-time analytics and hybrid OLTP workloads.

PayPal’s historic, multi-petabyte migration to BigQuery creates a unified, AI-ready data platform that accelerates real-time insights and enables gen AI–driven innovation across payments, fraud prevention, and customer experiences.

A concise overview of AWS's Large Model Inference container updates, highlighting LMCache-based long-context caching, EAGLE speculative decoding, expanded model support, and streamlined LoRA hosting to reduce latency and cost in large-scale LLM deployments.

Explores reinforcement fine-tuning (RFT) for Amazon Nova across Bedrock, SageMaker, and Nova Forge, detailing data preparation, reward design, and practical guidance for evaluation-based model customization.

A collaboration between Pacific Northwest National Laboratory and OpenAI to accelerate federal permitting through AI-driven workflows and regulatory analytics.

Nano Banana 2 speeds pro-level image generation and editing with real-time information, 4K upscaling, native text rendering, and enterprise-grade transparency via SynthID and C2PA, enabling seamless, scalable workflows in Vertex AI.

Explores Nano Banana 2, a Gemini 3.1 Flash Image model that fuses Nano Banana Pro's capabilities with Flash-speed generation, real-time web grounding, precise text rendering, and robust provenance across Google's ecosystem for high-quality visuals delivered at speed.

A technical overview of delivering Nasdaq eVestment data to the Databricks Lakehouse via Delta Sharing on Databricks Marketplace to enable real-time mandate discovery, AI-powered meeting prep, and automated workflow execution.

Deterministic automation via the Migration Intake and Processing Service (MIPS) replaces manual spreadsheets for Hyperforce migrations, enabling scalable, auditable decisioning across 95,000 organizations and 15,000+ requests with high-throughput auto-approval.