AI Agents: Less Capability, More Reliability, Please

The Power of Asymmetric Experiments @ Meta - Analytics at Meta - Medium by Analytics at Meta

• Asymmetric experiments make sense when (1) you don’t need to run many experiments in parallel, (2) recruiting people for the experiment is cheap, and (3) running the test intervention is expensive
• Asymmetric experiments make sense when your need for concurrent experiment capacity is low*.
• The test intervention is expensive. Asymmetric experiments have a smaller test group. This makes asymmetric experiments appealing when the cost of the test intervention is high. For example, if the test intervention has the possibility of negatively impacting the user experience, or would require an increase in compute costs.

Troubleshooting: The Skill That Never Goes Obsolete

• It’s easy to get lost in reactive problem whack-a-mole without stopping to think: what’s the real cause of this issue? What, exactly, is going on here?
• Writers are fond of saying that “writing is thinking”. Here are two ways I use writing as a troubleshooting tool:

Rubber duck debugging like a pro: I can often solve my problem by drafting a forum post without posting it. The effort required to articulate the salient details of the system and the problem, without looking dumb, is higher than the effort I have usually put in at the point I decide I need help. Corollary: making a forum post without sounding like I haven’t done my homework also tends to put me over my time/energy budget for solving a seemingly-trivial problem.

Behold the trail of crumbs: I find that writing and diagramming, while helpful for many troubleshooting projects, are essential for multi-session troubleshooting projects. I overestimate how much I will remember about the context, as well as how soon I will get around to continuing the project. A troubleshooting notes file, no matter how obvious or incomplete the information in it seems at the time I write it, leaves a trail of crumbs that I can follow next time. (I have often repeated, verbatim, an entire troubleshooting process, found the problem — and then remembered I troubleshot the exact system, and arrived at the same conclusion, years ago; but there was some hiccup, and I failed to order or install the new part.)

Build Bigger With Small Ai: Running Small Models Locally by MotherDuck

Who needs GitHub Copilot when you can roll your own AI code assistant at home

Amazon Nova expands Tool Choice options for Converse API - AWS

• Auto leaves tool selection entirely to Nova’s discretion, whether to call a tool or generate text instead. Auto is useful in use cases like chatbots and assistants where you may need to ask the user for more information, and is the current default. Any prompts Nova to return at least one tool call, from the list of tools specified, while allowing it to choose which tool to use. Any is particularly useful in machine to machine interactions where your downstream components may not understand natural language but might be able to parse a schema representation. Tool enable developers to request a specific tool to be returned by Nova. Tool is particularly useful in forcing a structured output by having a tool that has the return type as your desired output schema.

Reinventing notebooks as reusable Python programs by akshay, dylan, myles

Highlights from Git 2.49 by Taylor Blau

Preview: Amazon S3 Tables in DuckDB by Sam Ansmink, Tom Ebergen, Gabor Szarnyas

In S3 simplicity is table stakes by Dr Werner Vogels - https://www.allthingsdistributed.com/

• As the team started to look at scaling, they created a test account with an enormous number of buckets and started to test rendering times in the AWS Console — and in several places, rendering the list of S3 buckets could take tens of minutes to complete

fleetwood.dev

• Reading and writing from memory is extraordinarily slow when compared to computation
• Using multiple different approaches, we’ve derived a (non-exhaustive) list of design decisions that should hold for any AI inference accelerator, namely:

Hardware support for low precision data types Design for asynchronous transfers from day 1 Dedicated hardware for tensor aware memory transfers Replace your cache hierarchy with an outsized scratchpad for AI inference For a single accelerator, turn the memory bandwidth up to 11 Design for scale-out from day 1 Dedicated communication hardware should complement compute hardware

Performance of the Python 3.14 tail-call interpreter by Nelson Elhage

“Vibe Coding” vs Reality

Life Altering Postgresql Patterns

My Favorite Firefox Extensions

21 Unique Reasons Why Apache Hudi Should Be Your Next Data Lakehouse | Apache Hudi

Hard-Earned Lessons from a Year of Building AI Agents by Maya Murad

• LLMs can be harnessed for higher complexity problem-solving. By combining clever engineering with advanced prompting techniques, models could go beyond what few-shot learning could achieve. Retrieval-Augmented Generation (RAG) was an early example of how models could interact with documents and retrieve factual information. LLMs can also can dynamically interact with a designated environment via tool calling. When combined with chain-of-thought prompting, these capabilities laid the foundation for what would later be known as AI agents.

Resilience Best Practices: How Amazon Builds Well-Behaved Clients and Well-Protected Services

• Three operational strategies are suggested: load shedding, auto-scaling, and fairness.
• token bucket, leaky bucket, exponentially weighted moving average (EWMA), fixed window, or sliding window.
• To avoid making the situation worse for a dependency that is under stress, AWS suggests two patterns for well-behaved clients: circuit breakers, preventing the sustained overload of a dependency, and retries, letting the client retry every request up to N times using exponential backoff with jitter between requests

Prefix Aliases in SQL by Hannes Mühleisen

Gems of DuckDB 1.2 by The DuckDB team

• Starting with version 1.2.0, DuckDB supports OR and IN expressions for filter pushdown. This optimization comes especially handy when querying remote

Graph Databases after 15 Years – Where Are They Headed? by LDBC Linked Data Benchmark Council

In Praise of “Normal” Engineers

A 10x Faster TypeScript - TypeScript by Anders Hejlsberg

Introducing a SQL-based metrics layer powered by DuckDB by DuckDB

Ibis, DuckDB, and GeoParquet: Making geospatial analytics fast, simple, and Pythonic by DuckDB

GitHub - reloadware/reloadium: Hot Reloading and Profiling for Python by reloadware

Four steps toward building an open source community by Klint Finley

Is Apache Iceberg the New Hadoop? Navigating the Complexities of Modern Data Lakehouses by Ananth Packkildurai

• However, many data ingestion tools don’t natively support compaction, requiring manual intervention or dedicated Spark clusters.

AWS CDK Introduces Garbage Collection to Remove Outdated Assets

Mastering Spark: The Art and Science of Table Compaction

DuckDB goes distributed? DeepSeek’s smallpond takes on Big Data by mehdio

• Let’s recap the features of smallpond :Lazy evaluation with DAG-based execution – Operations are deferred until explicitly triggered.Flexible partitioning strategies – Supports hash, column-based, and row-based partitioning.Ray-powered distribution – Each task runs in its own DuckDB instance for parallel execution.Multiple storage layer options – Benchmarks have primarily been conducted using 3FS.Cluster management trade-off – Requires maintaining a compute cluster, though fully managed services like Anyscale can mitigate this.Potential 3FS overhead – Self-managing a 3FS cluster introduce significant additional complexity.

Announcing AWS Step Functions Workflow Studio for the VS Code IDE - AWS

Definite: Understanding smallpond and 3FS: A Clear Guide

• Smallpond’s distribution leverages Ray Core at the Python level, using partitions for scalability. Partitioning can be done manually, and Smallpond supports:

Hash partitioning (based on column values) Even partitioning (by files or row counts) Random shuffle partitioning

Data Products: A Case Against Medallion Architecture by Modern Data 101

A non-beginner Data Engineering Roadmap — 2025 Edition by Ernani Castro

Open Source Data Engineering Landscape 2025 - Apache DolphinScheduler - Medium by Apache DolphinScheduler

• Further consolidation in the open table format spaceContinued evolution of zero-disk architectures in real-time and transactional systemsQuest toward providing a unified lakehouse experienceThe rise of LLMOps and AI EngineeringThe expansion of the data lakehouse ecosystem in areas such as open catalog integration and development of native librariesThe increasing traction of single-node data processing and embedded analytics

macOS Tips & Tricks - saurabhs.org

The Commitment Inventory: Get More Done By Saying “No”

• Important: When the timer goes off, start the next item immediately, even if you haven’t finished. Otherwise, you’ll get distracted, and your attention will lag
• I’m now working within a 50-minute “burst,” and I have 9:56 left before I have to move on to a different checklist within Pitches, so I am cranking out this draft much faster and with more focus than I otherwise would.If your resistance is high, start with a short burst, then add five as you go up.  Forster recommends keeping it under 40, but when I’m writing, I use 50-minute bursts. Adapt to fit your needs. When you lose momentum, go back to 5 minutes. You can use bursts for your breaks, too. In 8 minutes, when I finish this burst, I will get a cup of coffee, stretch my legs, pet my dog Duke, and maybe read a quick news article.  Forster recommends being strict with the timer, or your attention will drift.

If it is worth keeping, save it in Markdown - Piotr Migdał by Piotr Migdał

Does Ibis understand SQL? – Ibis by Deepyaman Datta

Concurrency Control in Open Data Lakehouse | Apache Hudi

• cases where multiple writer jobs need to access the same table, Hudi supports multi-writer setups. This model allows disparate processes, such as multiple ingestion writers or a mix of ingestion and separate table service jobs to write concurrently.
• Apache Iceberg supports multiple concurrent writes through Optimistic Concurrency Control (OCC). The most important part to note here is that Iceberg needs a catalog component to adhere to the ACID guarantees. Each writer assumes it is the only one making changes, generating new table metadata for its operation. When a writer completes its updates, it attempts to commit the changes by performing an atomic swap of the latest metadata.json file in the catalog, replacing the existing metadata file with the new one.

Towards composable data platforms — Jack Vanlightly by Jack Vanlightly

• The OTFs introduce a new abstraction layer that can be used to virtualize table storage. The key is that it allows for the separation of data from metadata and shared storage from compute. Through metadata, one table can appear in two data platforms, without data copying. To avoid overloading data virtualization anymore, I will use the term Table Virtualization.
• The Modern Data Stack (MDS) failed to sustain itself. Few wanted to compose a data architecture from 10-15 different vendors. People want to choose a small number of trusted vendors, and they want them to work together without a lot of toil and headaches.

Common pitfalls when building generative AI applications by Chip Huyen

• Use generative AI when you don’t need generative AI
• Start too complex

Examples of this pitfall:

Use an agentic framework when direct API calls work. Agonize over what vector database to use when a simple term-based retrieval solution (that doesn’t require a vectordb) works. Insist on finetuning when prompting works. Use semantic caching.

• Forgo human evaluation

To automatically evaluate AI applications, many teams opt for the AI-as-a-judge (also called LLM-as-a-judge) approach — using AI models to evaluate AI outputs. A common pitfall is forgoing human evaluation to rely entirely on AI judges.

Redefining Data Engineering with Go and Apache Arrow by Thomas F McGeehan V

Emerging Patterns in Building GenAI Products

Mixture of Experts Explained

Microsoft Introduces CoRAG: Enhancing AI Retrieval with Iterative Reasoning

The Quest to Understand Metric Movements - Pinterest Engineering Blog - Medium by Pinterest Engineering

• root-cause analysis (RCA)
• Slice and DiceThis approach finds clues for a metric movement by drilling down on specific segments within the metric; it has found successes at Pinterest, especially in diagnosing video metric regressions
• General SimilarityIn this approach, we look for clues of why a metric movement happened by scanning through other metrics and finding ones that have moved very “similarly” in the same time period, whether in the same direction (positive association) or in the opposite direction (negative association).
• practice, we have found that the first two factors, Pearson and Spearman’s rank correlations, work best because:p-values can be computed for both, which help to gauge statistical significanceboth have more natural support for measuring negative associations between two time-seriesnon-monotonic (e.g. quadratic) relationships, for which Pearson and Spearman’s rank correlations won’t apply, don’t tend to arise naturally so far in our use-cases / time window of analysis
• Experiment EffectsThis third approach looks for clues of why metric movements happened by looking at what a lot of internet companies have: experiments.
• For each control and treatment group in an experiment, we perform a Welch’s t-test on the treatment effect, which is robust in the sense that it supports unequal variances between control and treatment groups. To further combat noise in the results, we filter experiments by each experiment’s harmonic mean p-value of its treatment effects over each day in the given time period, which helps limit false positive rates. We also detect imbalances in control and treatment group sizes (i.e., when they are being ramped up at a different rate from each other) and filter out cases when that happens.

You Should Use /tmp/ More

How to add a directory to your PATH by Julia Evans

• Bash has three possible config files: ~/.bashrc, ~/.bash_profile, and ~/.profile.

python-build-standalone now has Python 3.14.0a5 by Simon Willison

How to Use a Microphone

Scale Out Batch Inference with Ray

Systematically Improving RAG Applications - Jason Liu by Jason Liu

• 4 Re-Rankers¶ Instead of (or in addition to) fine-tuning a bi-encoder (embedding model), you might fine-tune a cross-encoder or re-ranker that scores each candidate chunk directly. Re-rankers can be slower but often yield higher precision. Typically, you do a quick vector search, then run re-ranking on the top K results.

Emerging Patterns in Building GenAI Products

• Self evaluation: Self-evaluation lets LLMs self-assess and enhance their own responses. Although some LLMs can do this better than others, there is a critical risk with this approach. If the model’s internal self-assessment process is flawed, it may produce outputs that appear more confident or refined than they truly are, leading to reinforcement of errors or biases in subsequent evaluations. While self-evaluation exists as a technique, we strongly recommend exploring other strategies.
• LLM as a judge: The output of the LLM is evaluated by scoring it with another model, which can either be a more capable LLM or a specialized Small Language Model (SLM). While this approach involves evaluating with an LLM, using a different LLM helps address some of the issues of self-evaluation. Since the likelihood of both models sharing the same errors or biases is low, this technique has become a popular choice for automating the evaluation process
• Human evaluation: Vibe checking is a technique to evaluate if the LLM responses match the desired tone, style, and intent. It is an informal way to assess if the model “gets it” and responds in a way that feels right for the situation. In this technique, humans manually write prompts and evaluate the responses. While challenging to scale, it’s the most effective method for checking qualitative elements that automated methods typically miss.
• However, embeddings are not ideal for structured or relational data, where exact matching or traditional database queries are more appropriate. Tasks such as finding exact matches, performing numerical comparisons, or querying relationships are better suited for SQL and traditional databases than embeddings and vector stores.

Why Most Machine Learning Projects Fail to Reach Production and How to Beat the Odds

Jujutsu VCS Introduction and Patterns by Kuba Martin

• While in Git you generally organize your commits in branches, and a commit that’s not part of a branch is scarily called a “detached HEAD”, in jj it’s completely normal to work on changes that are not on branches. jj log is the main command to view the history and tree of changes, and will default to showing you a very reasonable set of changes that should be relevant to you right now - that is (more or less) any local mutable changes, as well as some additional changes for context (like the tip of your main branch).

The Rise of Single-Node Processing: Challenging the Distributed-First Mindset by Alireza Sadeghi

The End of the Bronze Age: Rethinking the Medallion Architecture by

Bias and Fairness in Natural Language Processing - Thomson Reuters Labs - Medium by Navid Rekabsaz

The Emerging Role of AI Data Engineers - The New Strategic Role for AI-Driven Success by Ananth Packkildurai

Google Releases PaliGemma 2 Vision-Language Model Family by

Catching memory leaks with your test suite by Itamar Turner-Trauring

How the Apache Arrow Format Accelerates Query Result Transfer by Ian Cook, David Li, Matt Topol

ZenML VS Flyte VS Metaflow - MLOps Community by Ankur Tyagi

JavaScript Temporal is coming | MDN Blog by

Amazon (S3) Tables by Daniel Beach

Rill | Designing a Declarative Data Stack: From Theory to Practice by

The Hidden Cost of Over-Abstraction in Data Teams by Zakaria Hajji

Staff Engineer vs Engineering Manager by Alex Ewerlöf

PromptWizard: The future of prompt optimization through feedback-driven self-evolving prompts by Brenda Potts

Lessons Learned Implementing Metric Trees by Ergest Xheblati

Cost Optimized Vector Database: Introduction to Amazon OpenSearch Service quantization techniques | Amazon Web Services by

Apache Iceberg Won the Future — What’s Next for 2025? by Yingjun Wu

Building effective agents by

How I run LLMs locally by

Next.js 15.1 by

Hugging Face Smolagents is a Simple Library to Build LLM-Powered Agents by

A Pixel Parable by Facundo Olano

Hitting OKRs vs Doing Your Job by jessitron

Goodhart’s Law Isn’t as Useful as You Might Think by

Incremental Jobs and Data Quality Are On a Collision Course - Part 2 - The Way Forward — Jack Vanlightly by Jack Vanlightly

How AI is unlocking ancient texts — and could rewrite history by Marchant, Jo

Glue work considered harmful by

Goodbye Github Pages, Hello Coolify · Quakkels.com by

Tools Worth Changing To in 2025 by Matthew Sanabria

Databases in 2024: A Year in Review by

Dismantling ELT: The Case for Graphs, Not Silos — Jack Vanlightly by Jack Vanlightly

Things we learned about LLMs in 2024 by

Change query support in Apache Iceberg v2 — Jack Vanlightly by Jack Vanlightly

Collection of insane and fun facts about SQLite - blag by

Scalar and binary quantization for pgvector vector search and storage by Jonathan Katz

Turbocharge Efficiency & Slash Costs: Mastering Spark & Iceberg Joins with Storage Partitioned Join by Samy Gharsouli

Build Write-Audit-Publish pattern with Apache Iceberg branching and AWS Glue Data Quality | Amazon Web Services by

Designing data products by

On writing and getting from zero to done — Jack Vanlightly by Jack Vanlightly

Introducing AWS Glue Data Catalog automation for table statistics collection for improved query performance on Amazon Redshift and Amazon Athena | Amazon Web Services by

React v19 – React by

Tech predictions for 2025 and beyond by Dr Werner Vogels - https://www.allthingsdistributed.com/

First impressions of the new Amazon Nova LLMs (via a new llm-bedrock plugin) by

Use open table format libraries on AWS Glue 5.0 for Apache Spark | Amazon Web Services by

Migrating AWS Glue for Spark jobs to AWS Glue version 5.0 - AWS Glue by

Getting to Two Million Users as a One Woman Dev Team by

Why AI language models choke on too much text by Timothy B. Lee

AI-generated tools can make programming more fun by

The 150x pgvector speedup: a year-in-review by Jonathan Katz

How to generate unit tests with GitHub Copilot: Tips and examples by Greg Larkin

Introducing AWS Glue 5.0 for Apache Spark | Amazon Web Services by

Building Confidence: A Case Study in How to Create Confidence Scores for GenAI Applications - Spotify Engineering by alexandrawei

Storing times for human events by

DataFrames at Scale Comparison: TPC-H by

Enabling compaction optimizer - AWS Glue by

Top Python Web Development Frameworks in 2025 · Reflex Blog by

Building Python tools with a one-shot prompt using uv run and Claude Projects by

How to Speed Up Spark Jobs on Small Test Datasets by luminousmen

A high-velocity style of software development by

Building end-to-end data lineage for one-time and complex queries using Amazon Athena, Amazon Redshift, Amazon Neptune and dbt | Amazon Web Services by

A First Look at S3 (Iceberg) Tables by Nikhil Benesch

Amazon Bedrock Knowledge Bases now supports RAG evaluation (Preview) - AWS by

Amazon Aurora now available as a quick create vector store in Amazon Bedrock Knowledge Bases - AWS by

Why it took a long time to build that tiny link preview on Wikipedia by Jon Robson (David Lyall)

AWS Glue Data catalog now automates generating statistics for new tables - AWS by

Introducing AWS Glue 5.0 - AWS by

Table format comparisons - Change queries and CDC — Jack Vanlightly by Jack Vanlightly

Amazon S3 adds new functionality for conditional writes - AWS by

AWS Amplify introduces passwordless authentication with Amazon Cognito - AWS by

The Part of PostgreSQL We Hate the Most // Blog // Andy Pavlo - Carnegie Mellon University

What goes into bronze, silver, and gold layers of a medallion data architecture? | by Lak Lakshmanan | Sep, 2024 | Medium by Lak Lakshmanan

Using DuckDB-WASM for in-browser Data Engineering by Tobias Müller

Go talk to the LLM

The CDC MERGE Pattern. by Ryan Blue | by Tabular | Medium by Tabular

FireDucks : Pandas but 100x faster

#!/usr/bin/env -S uv run by

Amazon Data Firehose supports continuous replication of database changes to Apache Iceberg Tables in Amazon S3 - AWS by

Use data that looks like data - by Thorsten Ball by Thorsten Ball

Unit Tests As Documentation - by Teiva Harsanyi by Teiva Harsanyi

Anthropic’s upgraded Claude 3.5 Sonnet model and computer use now in Amazon Bedrock - AWS

• It will be fun to see how the computer use api will evolve

It’s Not Easy Being Green: On the Energy Efficiency of Programming Languages

Embeddings are underrated

Technology Radar | Guide to technology landscape | Thoughtworks

AWS announces a seamless link experience for the AWS Console Mobile App - AWS

init.py files are optional. Here’s why you should still use them | Arie Bovenberg

Talks - Reuven M. Lerner: Times and dates in Pandas by PyCon US

Talks - Bruce Eckel: Functional Error Handling by PyCon US

What’s New In Python 3.13 — Python 3.13.0 documentation

Visual Studio Code September 2024 by Microsoft

NotebookLM | Note Taking & Research Assistant Powered by AI

• Very interesting experiment from google which provides a summary of any article, video or file. Additionaly can generate a podcast talking about it!

Talks - Juliana Ferreira Alves: Improve Your ML Projects: Embrace Reproducibility and Production… by PyCon US

• These truly looks like a good improvement to an ml project 👀

Talks - Krishi Sharma: Trust Fall: Three Hidden Gems in MLFlow by PyCon US

Table format comparisons - Streaming ingest of row-level operations — Jack Vanlightly by Jack Vanlightly

Embeddings · Malaikannan

Copy-on-Write (CoW) — pandas 2.2.3 documentation

• Ptty big change to pandas 3.0 but one I think will bring a lot of clarity to data transformations

Announcing DuckDB 1.1.0 – DuckDB by The DuckDB team

Introducing Contextual Retrieval \ Anthropic

What is a Vector Index? An Introduction to Vector Indexing by Alejandro CantareroField CTO of AI, DataStax

Chunking · Malaikannan

Chronon - Airbnb’s End-to-End Feature Platform - InfoQ by Nikhil Simha

• This is a very high level overview of the feature platform but allowed me to get a better sense on why use this platform. But I’d love to test it to understand if for those without streaming I’m this isn’t a bit overkill

Deno 2.0 Release Candidate

Time spent programming is often time well spent - Stan Bright

What I tell people new to on-call | nicole@web

What is io_uring?

Memory Management in DuckDB – DuckDB by Mark Raasveldt

Yuno | How Apache Hudi transformed Yuno’s data lake

Google Proposes Adding Pipe Syntax to SQL - InfoQ by Renato Losio

PostgreSQL: PostgreSQL 17 Released!

The sorry state of Java deserialization @ marginalia.nu

• Interesting to see that duckdb is serving as the benchmark to process data in the order of GB’s

Llama 3.2: Revolutionizing edge AI and vision with open, customizable models

@daily_cache implementation in Python • Max Halford

Good programmers worry about data structures and their relationships by Engineer’s Codex

I Like Makefiles by Sebastian Witowski

• I tend to avoid make files because I hadn’t looked at how they worked and always associated either Java projects. I might want to take another look at them

Rethinking Analyst Roles in the Age of Generative AI by Ben Lorica 罗瑞卡

No, Data Engineers Don’t NEED dbt. | by Leo Godin | Jul, 2024 | Data Engineer Things by Leo Godin

Gotten my reading from +100 articles to 58! 🎉

Predicting the Future of Distributed Systems by Colin Breck

Continuous reinvention: A brief history of block storage at AWS | All Things Distributed by Werner Vogels

Iceberg vs Hudi — Benchmarking TableFormats | by Mudit Sharma | Aug, 2024 | Flipkart Tech Blog by Mudit Sharma

Splicing Duck and Elephant DNA by Jordan Tigani, Brett Griffin

How we sped up Notion in the browser with WASM SQLite by Carlo Francisco

Table format comparisons - How do the table formats represent the canonical set of files? — Jack Vanlightly by Jack Vanlightly

Should you be migrating to an Iceberg Lakehouse? | Hightouch by Hugo Lu

How data departments have evolved and spread across English football clubs - The Athletic by Mark Carey

How to use AI coding tools to learn a new programming language - The GitHub Blog by Sara Verdi

How to choose the best rendering strategy for your app – Vercel by Alice Alexandra MooreSr. Content Engineer, Vercel

Amazon’s Exabyte-Scale Migration from Apache Spark to Ray on Amazon EC2 | AWS Open Source Blog

Don’t Use JS for That: Moving Features to CSS and HTML by Kilian Valkhof by JSConf

AWS Chatbot now allows you to interact with Amazon Bedrock agents from Microsoft Teams and Slack - AWS

Astro 5.0 Beta Release | Astro by Erika

Making progress on side projects with content-driven development | nicole@web

AWS Glue Data Catalog now supports storage optimization of Apache Iceberg tables - AWS

Introducing OpenAI o1 | OpenAI

Rewrite Bigdata in Rust by Xuanwo

Recommending for Long-Term Member Satisfaction at Netflix | by Netflix Technology Blog | Aug, 2024 | Netflix TechBlog by Netflix Technology Blog

• Til: reward engineering. Measure proxy features and optimize for them to get to your actual end goal

We need to talk about ENUMs | boringSQL by Radim Marek

I spent 8 hours learning Parquet. Here’s what I discovered | by Vu Trinh | Aug, 2024 | Data Engineer Things by Vu Trinh

What I Gave Up To Become An Engineering Manager by Suresh Choudhary

Unpacking the Buzz around ClickHouse - by Chris Riccomini by Chris Riccomini

Introducing job queuing to scale your AWS Glue workloads | AWS Big Data Blog

”SRE” doesn’t seem to mean anything useful any more

Microsoft Launches Open-Source Phi-3.5 Models for Advanced AI Development - InfoQ by Robert Krzaczyński

Production-ready Docker Containers with uv by Hynek Schlawack

Many of us can save a child’s life, if we rely on the best data - Our World in Data by By: Max Roser

• Another article that I find eye opening when we look at data to improve our decisions

Why I Still Use Python Virtual Environments in Docker by Hynek Schlawack

• Suddenly this way of working, very similar to node modules, is being talked everywhere.

Python Developers Survey 2023 Results

• Was looking for this for a long time. Got to learn a bit more on twine, mlflow and sqlmodel. In terms of getting on track with python world I think I’m good with my current rss feed and podcast

Elasticsearch is Open Source, Again | Elastic Blog by ByShay Banon29 August 2024

Monitor data quality in your data lake using PyDeequ and AWS Glue | AWS Big Data Blog

• Data drift detection 👀
• Makes sense. Pydeequ is just a wrapper

How top data teams are structured by Mikkel Dengsøe

• As I work on a team that doesn’t have this kind of distribution I need to reflect a bit on the impact of being the sole data engineer on an ML team

Talks - Amitosh Swain: Testing Data Pipelines by PyCon US

uv: Unified Python packaging

Wow, need to test this out on my side projects.

CSS finally adds vertical centering in 2024 | Blog | build-your-own.org by James Smith

Timeless Skills For Data Engineers And Analysts by SeattleDataGuy

Skimmed the article but although high level I find the main points very true. Understanding the system, being on top of the state of the art. That and the tips for head of data

CSS 4, 5, and 6! With Google’s Una and Adam by Syntax

Great episode! Nice to see that css is getting better and better

I’ve Built My First Successful Side Project, and I Hate It by Sebastian Witowski

Wow, I’d love this one man saas but knowing how it can burn you out…

NodeJS Evolves by Syntax

Nice episode on the long sought features for node that have been introduced on bun and demo (single file, typescript support and top await async are the big ones for me)

Talks - Brandt Bucher: Building a JIT compiler for CPython by PyCon US

Python Insider: Python 3.13.0 release candidate 1 released

Great to see a new version coming along! Is pdb worth using with vs code? 🤔

Google kills Chromecast, replaces it with Apple TV and Roku Ultra competitor | Ars Technica by Samuel Axon

As a google to owner would be good to know that my working hardware wouldn’t brick just because I don’t want to upgrade

Can Reading Make You Happier? | The New Yorker by Ceridwen Dovey

Beyond Hypermodern: Python is easy now - Chris Arderne

Although I had my eyes already on rye I certainly didn’t know it was so full feature. Gotta try and change some of my projects to it ,

Visual Studio Code July 2024 by Microsoft

Python support slowly getting good

Introducing GitHub Models: A new generation of AI engineers building on GitHub - The GitHub Blog by Thomas Dohmke

Creativity Fundamentally Comes From Memorization

Gen AI Increases Workloads and Decreases Productivity Upwork Study Finds - InfoQ by Sergio De Simone

The report raises a good point about an increase in workload. With a big productivity boost bosses might start loading even bigger workloads than what we gained from the productivity boost

Ofc this isn’t good as people will feel overloaded

tea-tasting: a Python package for the statistical analysis of A/B tests | Evgeny Ivanov by Evgeny Ivanov

Super interesting IMO. Might give it a try on my team when we start deploying solutions to our clients

Data Science Spotlight: Cracking the SQL Interview at Instacart (LLM Edition) | by Monta Shen | Jul 2024 | tech-at-instacart by Monta Shen

Would be interesting to test this out. Have an example of a dataset that can be queried using duckdb. Given a question understand if a query is correct how to fix it and improve it’s performance. One in Sal and another in pyspark (or ibis/pandas)

Are you a workaholic? Here’s how to spot the signs | Ars Technica by Chris Woolston, Knowable Magazine

At the Olympics, AI is watching you | Ars Technica by Morgan Meaker, WIRED.comCrazy to see that now with ai it’s actually possible to survey an entire city either cameras

Use Apache Spark™ from Anywhere: Remote Connectivity with Spark Connect by Databricks

English sdk looks awesome but requires an OpenAI key. Could it be replaced with ollama?

Let’s Consign CAP to the Cabinet of Curiosities - Marc’s Blog by Marc Brooker

Interesting topics to research a bit more on CAP alternatives

Why Your Generative AI Projects Are Failing by Ben Lorica 罗瑞卡

In summary, to have a good AI product we need to have data with quality which requires good data governance. With this data we need to define useful products that we can measure it’s value using data driven metrics. We must also ensure the product has good practices avoiding security or bias issues

Engage your audience by getting to the point, using story structure, and forcing specificity – Ian Daniel Stewart

Resumed by:

Slack Conquers Deployment Fears with Z-score Monitoring - InfoQ by Matt Saunders

This is something I would love to implement. Allowing to define the metrics on which to evaluate a new feature, the expected hypothesis and revert the feature (i.e feature flags) automatically with a report on the experiment

DuckDB + dbt : Accelerating the developer experience with local power - YouTube

Could I replace Athena with this? I think the main blocker for me is I want to work with S3. And need to check how it runs for a really large dataset…

How Unit Tests Really Help Preventing Bugs | Amazing CTO

Good tip. For any project define the metric of code coverage goals and start increasing on the project

Mocking is an Anti-Pattern | Amazing CTO

Building an open data pipeline in 2024 - by Dan Goldin by Dan Goldin

Spark-Connect: I’m starting to love it! | Sem Sinchenko by Sem Sinchenko

This article wasn’t properly parsed by omnivore but the big takeaways:

• We can add plugins for extended functionality to our spark server
• Using spark connect we can implement any library in any language we want and send grpc requests to the server (spark connect server needs to be running)
• spark connect works on >3.5. Should be much better on v4.0
• If I truly want to be good in spark I eventually need to relearn scala/java
• Glue is cool but it’s still on version 3.3. All these goodies will take too long to be implemented in glue

So you got a null result. Will anyone publish it? by Kozlov, Max

After reading the statistics books I can see much more clear the value of proving a null hypophesis, this is the feeling I am getting out of the academia. We are seeing more research without any any added value. Goodhart’s law.

Maestro: Netflix’s Workflow Orchestrator | by Netflix Technology Blog | Jul, 2024 | Netflix TechBlog by Netflix Technology Blog

Sounds just like an airflow contender. with the plus of being able to run notebooks 🤔

How to Create CI/CD Pipelines for dbt Core | by Paul Fry | Medium by Paul Fry

Simplify PySpark testing with DataFrame equality functions | Databricks Blog by Haejoon Lee, Allison Wang and Amanda Liu

Good theme for a blog post on the changes of spark 4, this is really useful for human errors (been there multiple times)

How to build a Semantic Layer in pieces: step-by-step for busy analytics engineers | dbt Developer Blog by Gwen Windflower

So this will be generated on the fly as views by the semantic layer?, This looked neat until the moment I understood that the semantic layer requires dbt cloud

Meta’s approach to machine learning prediction robustness #### Engineering at Meta by Yijia Liu, Fei Tian, Yi Meng, Habiya Beg, Kun Jiang, Ritu Singh, Wenlin Chen, Peng Sun

Meta seems to be a couple of years ahead of the industry. The article doesn’t provide a lot of insights but gives a feeling of their model evaluation being mostly automated + having a good AI debug tool

Free-threaded CPython is ready to experiment with! | Labs

First step in a long way before we can get run python wihtout GIL. Interested on seeing if libraries like pandas will be able to leverage multithreading eventually with this

Amazon DataZone introduces OpenLineage-compatible data lineage visualization in preview | AWS Big Data Blog

Mixed feelings here. Great to see Open Lineage implemented at AWS. However it feels again that AWS just created the integration and won’t be driving the development of open lineage

Unlocking Efficiency and Performance: Navigating the Spark 3 and EMR 6 Upgrade Journey at Slack - Slack Engineering by Nilanjana Mukherjee

What could be improved to help this kind of migrations be done in a matter of days?, Livy might be deprecated in favor of spark connect. With their migration to Spark 3 and eventually 3.5 (not clear on this article) they could be interested in moving new jobs to connect , Basically solved issues by using the old behaviours. These will need to be migrated eventually. Would need to better understand these features , This looks like an important detail. With no explicit order spark can have random order of rows?, Cool to see these migrations and teams using open source solutions. EMR although expensive with a good engineering team can prove to be quite cost effective

DuckDB Community Extensions – DuckDB by The DuckDB team

Visual Studio Code June 2024 by Microsoft

The Rise of the Data Platform Engineer - by Pedram Navid by Pedram Navid

The need to define a data platform is something I see everywhere. It really looks like we are missing a piece here. Netflix maestro for example seems like a good contender for to solve the issue of (yet another data custom platform)

Lambda, Kappa, Delta Architectures for Data | The Pipeline by Venkatesh Subramanian

Write-Audit-Publish (WAP) Pattern - by Julien Hurault by Julien Hurault

This articles brings me the question. Can we improve dbt by using WAP? How does the rollback mechanism work when a process fails?

AWS Batch Introduces Multi-Container Jobs for Large-Scale Simulations - InfoQ by Renato Losio

Data Council 2024: The future data stack is composable, and other hot takes | by Chase Roberts | Vertex Ventures US | Apr, 2024 | Medium by Chase Roberts

Amazon DataZone now integrates with AWS Glue Data Quality and external data quality solutions | AWS Big Data Blog

Super interesting to see how we can enable data quality visibility

Pyspark 2023: New Features and Performance Improvement | Databricks Blog by in Industries

Cost Optimization Strategies for scalable Data Lakehouse by Suresh Hasundi

Good to use open data lakes showing the big cost and speed improvements

Prompt injection and jailbreaking are not the same thing

Flink SQL and the Joy of JARs by ByRobin MoffattShare this post

Catalogs in Flink SQL—Hands On by ByRobin MoffattShare this post

Catalogs in Flink SQL—A Primer by ByRobin MoffattShare this post

Why is this so hard? 😭