Notes from the people who build with open source

Benchmarks, systems deep dives, and hard-won lessons from OpenTeams engineers.

bootcArtifact Keeper
Infrastructure

Bare-Metal Rocky Linux Image Mode with Artifact Keeper

Deploy Rocky Linux image mode to bare metal with Artifact Keeper as the single source of truth for RPMs, signed bootc images, keys and Kubernetes workloads.

Brandon Geraci
Brandon Geraci
Senior Software Engineer at OpenTeams · · 8 min read
humanagent5515 min172.4 flakysame pixels as CI
AI Engineering

Reliable Visual Regression Testing for Humans and Coding Agents

Michał Krassowski
Michał Krassowski
·
agent✕ lint error✓ fixed
AI Engineering

Your AI Agent Ignored AGENTS.md. Your Linter Won't Let It.

Darshan Paudyal
Darshan Paudyal
·
Jev136 msLaya21 ms
AI Engineering

Where Does Jev Fit in a Software Supply Chain? We Benchmarked It Against the Open Alternatives

Brandon Geraci
Brandon Geraci
·
AI Engineering

Awesome Jupyter AI: A Map of 100+ Jupyter Extensions for AI

Michał Krassowski
Michał Krassowski
·

Latest articles

3 GiBof RAM
LLMs & Inference

MiniCPM5-2B quantization report

MiniCPM5-2B quantization report: the best GGUF weights and K/V cache quants on llama.cpp and BeeLlama.cpp, squeezing a SOTA model into 3 GiB RAM.

Guido Imperiale
Guido Imperiale
· 5 min read
summaryreviewedsurviveddeleted
AI Engineering

I Asked LLMs to Review Another LLM. They Still Got It Wrong

I tested LLM reviewer models on meeting summaries to see when they catch hallucinations, when they delete true claims, and how to choose one.

Khuyen Tran
Khuyen Tran
· 7 min read
$3.69vs$0.023
LLMs & Inference

LLMs: Intelligence vs. cost

LLM intelligence vs cost: logarithmic scales in plots hide the cost chasm between frontier and cheap models; actual OpenRouter prices alter the Pareto frontier.

Guido Imperiale
Guido Imperiale
· 6 min read
ubi9/ubi803 MB188 packages356 -> 16ubi-micro511 MB22 packages
Python Tooling

pixi on UBI-micro: A Safer, Smaller Multi-Stage Container Build

How a UBI-micro multi-stage build shrinks a pixi container's attack surface and CVE count, with side-by-side Trivy and Grype scans on UBI full, minimal, and micro.

Brandon Geraci
Brandon Geraci
· 12 min read
40 commentsthe 3 that matter
Practices

The Best Code Review Says Less

Far too many AI code review comments are noise. Pair a human editor with AI, surface only the findings that matter, and encode that discipline in a reusable skill.

Johnny Bouder
Johnny Bouder
· 5 min read
driveridestochastic, not deterministic
AI Engineering

You Don't Drive AI. You Ride It.

AI is stochastic, not deterministic. You don't drive it, you ride it: make documents the memory, use the variation to prototype, and verify every public claim.

Amelia (Mia) Thurdekoos
Amelia (Mia) Thurdekoos
· 11 min read
written by AIread by humans
Practices

Who Is Your Code's Audience? An Engineering Team Talks AI-Written Code

Learn why your code's audience decides how much readability matters, as an engineering team debates AI-written code, review, and accountability.

Amelia (Mia) Thurdekoos
Amelia (Mia) Thurdekoos
· 9 min read
a + bPyNumberMethodsnb_addnb_subtractnb_multiply
Python Tooling

The tp_as_number Slot and Binary Operation Dispatch in CPython

How CPython dispatches binary operations via tp_as_number slots, and how TorchDynamo mirrors that model to improve correctness and reduce ad-hoc special cases.

Guilherme Leobas
Guilherme Leobas
· 6 min read
doclingmarkerllamaparse
AI Engineering

PDF Table Extraction: Docling vs Marker vs LlamaParse Compared

Compare three Python tools for PDF table extraction: Docling, Marker, and LlamaParse. Learn which handles merged cells and multi-level headers best.

Khuyen Tran
Khuyen Tran
· 10 min read
localLLMsandboximport toolsrun(task)# no network
LLMs & Inference

Sandboxing Code Mode for Local LLM Agents

Code mode can make local LLM agents more practical, but executing model-written code brings sandboxing back into the architecture.

Nick ByrneKhuyen Tran
Nick Byrne & Khuyen Tran
· 4 min read
plugin.tsAI draftwriteloadtestrefine
AI Engineering

Plugin Playground AI Integration for Faster Plugin Prototyping

Learn how we integrated AI into Plugin Playground to help you create, edit, test, package, and share JupyterLab plugins faster in JupyterLite and Binder.

Anuj Kumar Singh
Anuj Kumar Singh
· 5 min read
deliberaterushedbetter code
Practices

Slow Down — Simple Lessons for Guiding AI and Shipping Better Code

Practical lessons for shipping better code, staying in control, keeping your skills sharp, and getting real value from AI coding tools without losing yourself in the hype.

Johnny Bouder
Johnny Bouder
· 3 min read
env recreate, same ML projectuv2.1spixi21.8spoetry86.6smamba104.8spip163.5sconda163.6s
Python Tooling

We Benchmarked 6 Python Package Managers on a Real ML Project. Here's What We Found.

Head-to-head benchmark of pixi, uv, conda, mamba, pip, and poetry on a real ML/AI project with 25+ mixed conda-forge and PyPI dependencies.

Brandon Geraci
Brandon Geraci
· 6 min read
$ llama-server  qwen3-coderlocal · no cloudevals100% pass != prod-ready
LLMs & Inference

What I Learned Making a Local LLM Do Real Work

Learn what makes a local LLM agent reliable enough for real work: evals, deterministic Python logic, and why a better model beats engineering.

Adam Lewis
Adam Lewis
· 7 min read
SKILL.mda text filewhile task: plan -> act ✓ keeps statean agent
AI Engineering

From Skill to Agent: When a Text File Isn't Enough

When does a Claude Code skill stop being enough? See why credential security pushes real workflows toward proper agent architectures.

Adam Lewis
Adam Lewis
· 5 min read

Never miss a post

Benchmarks, deep dives, and lessons from OpenTeams engineers. Leave your email, or follow the feed.

Or subscribe via RSS