ARTICLES
NVIDIA Says Small Language Models Are The Future of Agentic AI | SLM | 5 min | Cobus Greyling | Personal Blog
Small models are faster, safer, and better suited for real-time AI. NVIDIA explains why they may outpace large LLMs in practical applications.
Ververica’s Foolproof Path to AI: Why Streaming ETL Fuels Next-Gen Machine Learning | 5 min | MLOps | Ben Gamble | Ververica Blog
Streaming pipelines reduce latency, improve feature freshness, and unlock continuous model updates.
What Every AI Engineer Should Know About A2A, MCP & ACP | 7 min | AI | Edwin Lisowski | Personal Blog
A practical guide to three core agentic system patterns for reasoning and structured control.
TUTORIALS
How to Profile Models in PyTorch | 5 min | MLOps | Quentin-Anthon | Personal Blog
Learn how to trace, debug, and optimize model performance using PyTorch’s native tools.
Setting up a local Langfuse server with Kubernetes to trace Agentic systems | 5 min | LLMOps | Jetze Schuurmans | Xebia Blog
Walkthrough for running Langfuse locally to trace and debug LLM agents with full control.
Large Language Models as Classification Engines: Overkill, or Awesome? | 12 min | LLM | Katherine Munro | Towards AI Blog
A comparison of LLMs versus traditional classifiers in terms of cost, performance, and practicality.
Handling Long-Running Operations in Microsoft Fabric REST API | 3 min | Data Engineering | Microsoft Ignite Blog
How to manage async operations in Fabric using polling and status tracking.
TOOL
Embedding User-Defined Indexes in Apache Parquet Files | 7 min | Data Engineering | Qi Zhu, Jigao Luo, Andrew Lamb | Apache DataFusion Blog
DataFusion introduces custom Parquet indexing for faster queries on large datasets.
DATA LIBRARY
MedGemma Technical Report | LLM | Hugging Face
New findings suggest long-context LLMs may be overrated for many tasks, and retrieval methods often perform better.
DATA TUBE
The Agent Factory - Episode 2: Multi-Agent Systems, Concepts & Patterns | 23 min | Gen AI | Vlad Kolesnikov, Shir Meir Lador | Google Cloud Tech
Vlad Kolesnikov and Shir Meir Lador explain how to design collaborative agents using swarms, supervisors, and context engineering.
PINNACLE PICKS
Your last week top picks:
Direct Data Sharing using Delta Sharing - Introduction: Our Journey to Empower Partners at Zalando | Data Governance | 5 min | Lokeshbabu Radhakrishnan | Zalando Engineering Blog
Zalando is rolling out Delta Sharing to give partners real-time, governed access to data. No more manual exports, just scalable interoperability across teams and systems.
Introducing DataFrame API Support for Table-Valued Functions in Databricks | 5 min | Data Frames | Allison Wang, Takuya Ueshin, Jules Damji | Databricks Blog
You can now reuse complex logic with parameterized TVFs directly in the DataFrame API. Write cleaner pipelines without losing SQL-style reusability.
Why We Replaced Kafka with gRPC for Service Communication | 5 min | Data Engineering | Himanshu Singour | Personal Blog
One team ditched Kafka in favor of gRPC to reduce latency and simplify infra. A thoughtful case study that challenges default architectural choices.
________________________
Have any interesting content to share in the DATA Pill newsletter?
