Principia — Applied AI notes from Magellan Partners

We build, measure and explain applied AI — and publish what we learn.

Field notes, teaching material, internal research, benchmarks and live demos from the Magellan Partners AI team. Written by the people who ran the experiments.

Start reading Subscribe (RSS)

Lab status · Sept 2026

Articles published8

Benchmarks (reproducible)1

Demos live1

Last update2 Sept 2026

Latest

All articles →

An annotated printed copy of the AI Act on a desk

Watch · Policy

The AI Act's GPAI obligations, read by people who have to comply with them

C. Renard · 2 Sept · 1 min

Watch · LLM

Speculative decoding is finally boring — and that's good news

A. Martin · 1 Sept · 1 min

Learn · RAG

Chunking, explained with a ruler and a pair of scissors

S. Dubois · 28 Aug · 3 min

Research · Evaluation

Can a judge model grade its own family? Preliminary results

V. Levy dit Vehel · 25 Aug · 1 min

Watch · Agents

MCP one year on: what actually got adopted

K. Nguyen · 22 Aug · 1 min

Demos · RAG

New: the chunking playground now supports scanned PDFs

S. Dubois · 20 Aug · 1 min

Themes

Agents LLM RAG Evaluation MLOps Classical ML Data Tooling Safety Policy

Featured

Benchmarks · Agents

Eleven open-weight models, one tool-calling harness, 2,200 tasks

The ranking is not the one you expect. Smaller models with a strict JSON schema beat larger ones with free-form tool calls — and retries, not raw capability, explain most of the gap.

Victor Levy dit Vehel, Alice Martin · 2 Sept 2026 · 3 min · dataset · repo

Learn · RAG

Chunking, explained with a ruler and a pair of scissors

Before embeddings, before vector databases, there is a much dumber question: where do you cut the document? It turns out the answer decides most of your retrieval quality.

S. Dubois · 28 Aug · 3 min

Research · Evaluation