Andrey Buzin

talking to robots all day @vercel | ▲ uv add ai

[Andrey Buzin @_abuzin]

@MelkeyDev oh finally

[Andrey Buzin @_abuzin]

@mrloldev @ctatedev let me know what features you look forward to the most! we're definitely working on making it as expressive as the ts sdk

[Andrey Buzin @_abuzin]

@mrloldev @ctatedev it's not a port of the typescript version, and it's diverging from it in some fundamental ways. we do plan to expand _interoperability_ with it though, that's a big focus from the start

[Andrey Buzin @_abuzin]

@ctatedev Thank you so much for the shout out!

[Andrey Buzin @_abuzin]

@m0hilll @cramforce making it durability-friendly is also a huge focus

[Andrey Buzin @_abuzin]

@m0hilll @cramforce we’re trying to keep it very intuitive and easy to customize, and close to plain async python.

[Chris Tate @ctatedev]

AI SDK for Python now available

Agent loops, tool calling, streaming

uv add ai

[Malte Ubl @cramforce]

uv add ai
https://t.co/bIDZvxxWMt

[Andrey Buzin @_abuzin]

@1st1 why is Sully looking like he’s embarrassed? 🙈

[Andrey Buzin @_abuzin]

@andrewqu Let me give it to you straight, there's one wart in the canonical substrate.

[Andrey Buzin @_abuzin]

@mamuso @cursor_ai huge for cursor! 🐐

[Andrey Buzin @_abuzin]

@MelkeyDev your token usage kinda weak ¯\_(ツ)_/¯

[Andrey Buzin @_abuzin]

@1st1 at least you didn’t blueball it

[Gel @geldata]

In our new blog post, Andrey shows how to use Gel’s unique features to simplify and streamline the FastAPI + Pydantic AI stack for adding long-term memory to agents.

https://t.co/LOXGxDq18a

[Gel @geldata]

New blog post about Knowledge Graphs. This is part 1, where we talk about the core principals and different approaches.

[Andrey Buzin @_abuzin]

@solarperimeter @geldata Of course! We have an entire extension dedicated to that: https://t.co/BNsGUIDX57. The embeddings bit is fully automatic, although there might be a learning curve to using it (we’re working on that)

[Andrey Buzin @_abuzin]

Nothing quite as anxiety inducing as having a model edit 2000 notes you've gathered over the past few years, and just... trusting it not to mess up 🙈

https://t.co/wE2RrYAFZe

[Andrey Buzin @_abuzin]

@jerryjliu0 Proud to have been a part of this! We built this to be a part of our internal RAG applications, where using @llama_index is a no-brainer. But personally I think @MathpixApp is kind of a hidden gem in this, indispensable when trying to get anything done with a LaTeX-born PDF.

[Jerry Liu @jerryjliu0]

Building Multimodal RAG over Scientific Papers 🔬📑

Excited to partner with the Tensorsense team (@_abuzin, Sergey, Mark) on a cookbook that shows you how to properly parse, index, and retrieve over scientific papers - this includes diagrams, figures and math equations!

1️⃣ Convert PDF to Latex with @MathpixApp
2️⃣ Process embedded objects with a multimodal model
3️⃣ Retrieve and query over a multimodal collection of items

This gives you the ability to have the LLM understand all the complex sources in your doc: text, images, and tables.

Check it out (first image is example table, second/third images are notebook screenshots): https://t.co/VmWKkkM2hB

[Andrey Buzin @_abuzin]

@younglazybreech Хотя бы не грязный коврик у двери…