Autograd and Mutation

How does PyTorch autograd deal with mutation? In particular, what happens when a mutation occurs on a view, which aliases with some other tensor? In 2017, Sam Gross implemented support for in-place operations on views, but the details of which have never been described in plain English… until now.

https://blog.ezyang.com/2026/03/autograd-and-mutation/

ATLAS

Adaptive Test-time Learning and Autonomous Specialization.

https://github.com/itigges22/ATLAS

From zero to a RAG system: successes and failures

Building a production RAG system is far more about data pipelines, indexing strategy, and infrastructure tradeoffs than model choice, with most failures coming from scaling, retrieval quality, and compute constraints. The key lesson is that RAG success depends on iterative engineering and system design discipline, not just plugging in an LLM, with real-world performance shaped by bottlen...

https://en.andros.dev/blog/aa31d744/from-zero-to-a-rag-system-successes-and-failures/

OpenSpace

Make Your Agents: Smarter, Low-Cost, Self-Evolving.

https://github.com/HKUDS/OpenSpace

Timesliced reservoir sampling: a new(?) algorithm for profilers

Reservoir sampling lets you pick a sample from an unlimited stream of events; learn how it works, and a new variant useful for profilers.

https://pythonspeed.com/articles/reservoir-sampling-profilers/

Making friendly classes

A friendly class accepts sensible arguments, has a nice string representation, and supports ...

https://www.pythonmorsels.com/friendly-classes/

Enforce Business Logic in the Database with Django

Use transactions, locks, and constraints in Django to enforce business logic in the database.

https://lincolnloop.com/blog/enforce-business-logic-in-the-database-with-django/

Components of A Coding Agent

The post breaks down coding agents into a few core parts: a model, tools, instructions, and a loop for planning, acting, and verifying. Its main point is that good coding agents are less about raw model power and more about how well they coordinate context, tool use, and iterative feedback.

https://magazine.sebastianraschka.com/p/components-of-a-coding-agent

caveman

A Claude Code skill/plugin and Codex plugin that makes agent talk like caveman - cutting ~75% of output tokens while keeping full technical accuracy. Plus a companion tool that compresses your memory files to cut ~45% of input tokens every session.

https://github.com/JuliusBrussee/caveman

GuppyLM

A ~9M parameter LLM that talks like a small fish.

https://github.com/arman-bd/guppylm

profiling-explorer

Table-based exploration tool for Python profiling data (pstats files).

https://github.com/adamchainz/profiling-explorer

code-review-graph

Local knowledge graph for Claude Code. Builds a persistent map of your codebase so Claude reads only what matters - 6.8× fewer tokens on reviews and up to 49× on daily coding tasks.

https://github.com/tirth8205/code-review-graph

void-model

An open-source video editing model that removes objects from footage and also models the scene’s resulting physical interactions, not just the missing pixels.

https://github.com/Netflix/void-model

I Replaced Kafka, Redis, and RabbitMQ With One Tool. Here’s What I Learned.

The post’s core idea is that NATS JetStream can replace separate Kafka, Redis, and RabbitMQ setups by covering streaming, queueing, and pub/sub in one system. It argues this can simplify architecture and reduce the operational overhead of running multiple messaging tools for different workloads.

https://scalebites.substack.com/p/i-replaced-kafka-redis-and-rabbitmq

dj-signals-panel

Display registered Django signals and receivers, showing what fires and where.

https://github.com/yassi/dj-signals-panel

Decentralized AI in 50 Lines of Python

A peer-to-peer AI that answers your friends' messages using local data and privacy controls.

https://iamtrask.github.io/2026/04/07/decentralized-ai-in-50-lines/

Overglade-Badges

Zero power PCB hackathon badges.

https://github.com/KaiPereira/Overglade-Badges