engblogs

Omada Health scales personalized nutrition coaching by fine-tuning Llama 3.1 on Amazon SageMaker AI using QLoRA, enabling HIPAA-compliant, real-time nutrition education with LangSmith-based evaluation.

MANZANO is a simple, scalable unified multimodal model that couples a hybrid image tokenizer with a shared vision encoder and dual adapters to support continuous image-to-text understanding and discrete text-to-image generation within a single autoregressive LLM, aided by an auxiliary diffusion decoder that translates image tokens to pixels, all trained under a unified recipe to achieve state-of-the-art results with minimal task conflicts.

Unified crossmodal search enabled by a single Amazon Nova Multimodal Embeddings model that maps text, images, audio, video, and documents into one shared vector space for end-to-end ecommerce retrieval using cosine similarity, S3 Vectors, and Bedrock-backed embeddings.

Vera Rubin NVL72 racks join Lambda's Superclusters, delivering a 72-GPU NVLink domain to enable production-scale AI with model-parallel training and MoE-powered inference at higher efficiency.

A technical walkthrough of how 7-Eleven replaced scattered maintenance documents with a Databricks Agent Bricks powered AI assistant, integrating Unity Catalog, vector search with Embeddings Compute, and a Teams Bot to surface contextual manuals, diagrams and images in seconds while reducing downtime.

Unifying GenAI on Databricks to deliver a safe, scalable home-services platform through a hybrid CNN-LLM workflow that enhances privacy, trust, and collaboration across Thumbtack’s data science and engineering stack.

Explores scaffolds and prototyping tools for shaping the user experiences of interface agents, via AgentBuilder design probes, requirements elicitation, and in-situ on-device multi-agent prototyping.

AdaBoN introduces a prompt-adaptive Best-of-N alignment framework featuring a two-stage algorithm: an initial exploratory phase to estimate per-prompt reward distributions, then adaptive budget allocation to improve LM-RM alignment while reducing inference latency across diverse prompts.

A concise technical overview of the cross-sector partnership between OpenAI, SoftBank Group, and SB Energy to explore AI-powered energy technologies.

Replace read-only ephemeral sandboxes with Sprites—durable, instantly bootable computers that support checkpoint/restore, ample storage, and global Anycast access for seamless development-to-prod workflows.

An end-to-end guide to accelerating LLM inference via post-training weight-and-activation quantization using AWQ and GPTQ on Amazon SageMaker AI, reducing memory and latency without retraining.

Introducing the Snorkel Agentic Coding Benchmark—a real-world, end-to-end evaluation suite for AI coding agents that spans multiple languages, 100 multi-step tasks across four difficulty tiers, and rigorous long-horizon planning, error recovery, and sandboxed execution.