LLMs from Scratch – Practical Engineering from Base Model to PPO RLHF

This video provides a hands-on guide to building a large language model entirely from scratch in PyTorch, covering every step from core transformer design to advanced alignment with RLHF. By the end, viewers gain practical experience in implementing, training, scaling, and aligning their own custom LLMs.

https://www.youtube.com/watch?v=p3sij8QzONQ

Unlocking Performance in Python's Free-Threaded Future: GC Optimizations

A description of the performance optimizations made to the free-threaded garbage collector for Python 3.14.

https://labs.quansight.org/blog/free-threaded-gc-3-14

Air

The new web framework that breathes fresh air into Python web development. Built with FastAPI, Starlette, and Pydantic.

https://github.com/feldroy/air

onyx-dot-app / onyx

Open Source AI Platform - AI Chat with advanced features that works with every LLM

https://github.com/onyx-dot-app/onyx

Helium

Private, fast, and honest web browser.

https://github.com/imputnet/helium

memvid

Video-based AI memory library. Store millions of text chunks in MP4 files with lightning-fast semantic search. No database needed.

https://github.com/Olow304/memvid

DuckDB vs Polars. Wait. DuckDB and Polars.

The article emphasizes that DuckDB and Polars are not direct competitors but complementary tools in the Modern Data Stack, with each excelling in different contexts: DuckDB is best for SQL-heavy analytics and embedding as a query engine, while Polars suits end-to-end ETL pipelines and DataFrame-centric workflows. The choice depends on your problem context, team comfort, and use case rath...

https://www.confessionsofadataguy.com/duckdb-vs-polars-wait-duckdb-and-polars/

search_evals

Batteries-included eval framework for search APIs.

https://github.com/perplexityai/search_evals

Simplifying Resource Management in mssql-python through Context Manager

The article introduces context manager support in the mssql-python driver, allowing Python applications to manage SQL Server and Azure SQL resources more safely and efficiently using Python's "with" statement. This feature automates opening and closing of connections and cursors, as well as commit and rollback of transactions, reducing boilerplate code, preventing resource leaks, and ens...

https://devblogs.microsoft.com/python/simplifying-resource-management-in-mssql-python-through-context-manager/

bytedance / Dolphin

The official repo for “Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting”, ACL, 2025.

https://github.com/bytedance/Dolphin

Effective context engineering for AI agents

Context is a critical but finite resource for AI agents. In this post, we explore strategies for effectively curating and managing the context that powers them.

https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents

Building an AI agent from scratch in Python

How to implement a single AI agent with an LLM API and no frameworks.

https://www.leoniemonigatti.com/blog/ai-agent-from-scratch-in-python.html

I stopped using the Django shell because of marimo

The video explains how marimo, a modern Python notebook, enhances productivity in Django projects by integrating SQL support, coding agents, and interactivity beyond the traditional Django shell. It demonstrates using marimo for database connections, asynchronous Django ORM operations, custom admin interfaces, and reactive, interactive notebooks that combine SQL, Python, and UI elements ...

https://www.youtube.com/watch?v=ksGIAy2s_5k