engblogs

A concise technical analysis of OpenAI's partnership with Cerebras, examining implications for AI model training, hardware acceleration, and scalable compute architectures.

Integrating the User True Interest Survey (UTIS) with Facebook Reels' large-scale ranking to turn real-time user feedback into more personalized, diverse, and high-quality video recommendations while addressing data sparsity and bias.

Explores how Algonomy's RetailAI, built on the Databricks platform, fuses AI-first retail algorithms, agentic AI, and scalable data governance to enable rapid, outcome-driven planning across demand forecasting, replenishment, and promotions.

BigQuery's Comments to SQL enables vibe querying by embedding natural language expressions directly in SQL, translating them into executable queries to speed up complex data analysis.

Automates the end-to-end creation of customer intelligence documents (DOR) via agentic design on Google Cloud, leveraging Vertex AI, RAG, Gemini models, and the ADK-based agent orchestration for scalable, grounded results.

A practical case study of how a non-CUDA developer used Mojo to outpace a CUDA NF4 dequantization kernel by using memory-efficient packed stores, occupancy tuning, and AI-assisted experimentation.

A deep dive into how a memory-optimized DNS response merged CNAMEs after A/AAAA records in 1.1.1.1, triggering resolution failures and exposing RFC ambiguities around RRsets, CNAME ordering, and the need for an explicit standard.

A technical overview of Duolingo's Video Call with Falstaff, a real-time, coaching-driven video feature that uses adaptive prompts and multilingual feedback to guide speaking practice and build confidence.

Databricks Genie powers natural-language, self-service analytics in Atlassian Rovo, democratizing data access and enabling real-time, data-driven insights across teams.

Scaling unit test coverage at The New York Times with generative AI tooling, a guarded two-loop workflow, and prompts to safely expand tests across a large monorepo.

NYT demonstrates scalable unit-test coverage via AI-assisted test generation in a monorepo, using structured prompts and human supervision to hit 80%+ coverage while managing risk.

AI-powered unit-test generation scales test coverage across The New York Times' News monorepo, balancing automated testing with human oversight to speed feature development.