Matt Green

the amount of coffee I'm drinking why agents write my code is giving me the jitters

every time I write snippets it's a joy, every time I try to import a module, set up a new codebase, or host a project it's a terror

DataSRE.ai - Intelligent Spark Infrastructure Management Save up to 50% on your Spark compute bills with AI-driven optimization, intelligent autoscaling, and automated infrastructure management.

Just launched a free AI powered Spark Optimizer that examines spark logs to make cost and efficiency recommendations!

Looking for feedback from Spark users #databs

datasre.ai

do people here still care about tech or we just focused on the tariff news too?

mcp-leftpad An MCP server that exposes left-pad as a tool. Latest version: 1.0.0, last published: 6 days ago. Start using mcp-leftpad in your project by running `npm i mcp-leftpad`. There are no other projects in...

Ever wish your AI agent had more external dependencies?

announcing mcp-leftpad!

Using the power of MCP, AI can now easily leftpad strings

www.npmjs.com/package/mcp-...

"builds character"

GitHub - emgeee/mcp-ollama: Query model running with Ollama from within Claude Desktop or other MCP clients Query model running with Ollama from within Claude Desktop or other MCP clients - emgeee/mcp-ollama

spent a few hours building an mcp server to connect ollama models to mcp clients (like claude desktop)
github.com/emgeee/mcp-o...

been getting into agents a lot more recently and I've been struggling to define just what they mean. Best I can come up with is LLMs + tool usage executing in a dynamic compute graph. Sound right?

Hard agree

One fun bit about this project was I was able to use the claude desktop app to write the docker/docker compose files without copy/pasting

GitHub - feast-dev/feast-denormalized-tutorial: Feast + Denormalized Feast + Denormalized. Contribute to feast-dev/feast-denormalized-tutorial development by creating an account on GitHub.

Just published a new example of using the Denormalized Stream processing engine to compute real-time fraud features and sink to a Feast datastore - blog post coming out soon! github.com/feast-dev/fe...

I hear claude will write a mean operator for you

Open web-ui is building an open source llm ui— I’ve actually set this up and highly recommend!

GitHub - elixir-nx/nx: Multi-dimensional arrays (tensors) and numerical definitions for Elixir Multi-dimensional arrays (tensors) and numerical definitions for Elixir - elixir-nx/nx

NX is trying to bring Ai to elixir with python model interop github.com/elixir-nx/nx

@mozilla.ai builders day has begun!

I'm really surprised by this interaction, what did people think an open protocol meant?

definitely noticed this too

Posthog has a half-decent free tier

Finally got around to playing with @anthropic.com MCP protocol and it's very well done. Seems like a step in the right direction and I hope it continues to gain traction

anytime someone asks me how they should get started developing with LLMs

denormalized Embeddable stream processing engine

just dropped denormalized 0.0.10
- (fix): engine no longer panics when handling really late data
- (feat) default to using the kafka timestamp if no timestamp column is specified

pypi.org/project/deno...

How Do Merkle Trees Work? | Baeldung on Computer Science A quick and practical guide to Merkle trees.

Merkle Trees are such a powerful data structure for distributed systems. They can allow you save a lot of bandwidth for the small cost of re-computing hashes of data. They also are what allow you to trust data retrieved from untrusted sources www.baeldung.com/cs/merkle-tr...

honestly thought "databs" was "data bull-shit"

agree, though sometimes models that are deployed are also useless (or worse, less than useless)

I've actually used this to great success when trying to log food. Take a picture of the menu -> select text description of food -> paste into food logging AI description box. Worked decently well for simply/medium complexity dishes

Seeing as Apache manages a lot of important data projects, #databs might be interested in this post

With all the ways ppl communicate it feels even more confusing to get help

Have problem with a project?

go to the Github -> README suggests you ask question in discord -> discord ppl tell you to open issue on github -> bug fix merged, subscribe to mailinglist for release notification in 1 month

The part about mailing lists is also spot on.

Hot take: I never understood how to read and interact with mailing lists and that confusion discouraged me greatly from contributing more early in my career.

Skill issue? Perhaps

Github made it much simpler and more welcoming

Fantastic post.

When I started getting involved with the DataFusion project I never understood why it would take a month for my contributed changes to be released. It meant I basically had to maintain my own fork and build my project against that -- a process that become rather burdensome

playing with the bsky firehose and it seems that record timestamps are in an inconsistent format. Is there some library that can properly parse these into unix time automatically?