Niccolò Ridi

~900 downloads in the first few days on PyPI – but only ~10% look like real Python environments (the rest are mirrors / bots).

So roughly 20–30 genuine installs per day.

For a niche CJEU empirical toolkit, I’ll take that.

pypi.org/project/cjeu...

Thank you!

6/ MIT licensed, on PyPI now. Feedback, issues, and contributions welcome. pypi.org/project/cjeu-py

5/ Full CLI with search built in – query your local data by full text, party name, citation graph, subject matter, or legislation. Or hit CELLAR directly with headnote search, no local data needed.

4/ Everything is designed for standard research workflows: Parquet files at every stage, pandas DataFrames in and out, resumable downloads, deterministic caching. Each pipeline step is independently inspectable and re-runnable.

3/ I'm most pleased with the interactive citation network export. Nodes sized by PageRank, coloured by community/procedure/year, filterable by subject matter, court, and date. Self-contained HTML – just open in a browser. Also exports GEXF for Gephi.

2/ Optional LLM classification sorts each citation by precision, use, treatment, and topic – taxonomy drawn from Marc Jacob's work on precedent at the CJEU. Works with Gemini out of the box, or any local model via Ollama/vLLM/LM Studio.

1/ The pipeline: download case metadata + full texts from the EU's official CELLAR endpoint (no scraping), parse XHTML headers for court composition, parties, and representatives, extracts case-law citations with para-level context across 14 regex patterns + italic markers + party name matching.

cjeu-py Python toolkit for empirical research on the Court of Justice of the European Union

🧵 I just released cjeu-py – an open-source Python toolkit for empirical research on the Court of Justice of the EU.

pip install cjeu-py

It collects structured data from CELLAR, parses judgments, extracts citations, and optionally classifies them with an LLM. Thread 👇

pypi.org/project/cjeu...

This study identifies current GenAI strengths & limitations in legal argumentation and critically informs:

➡️ Best practices in prompt engineering
➡️ Human-AI collaboration strategies
➡️ Emerging regulatory policies for legal education & practice

Read it here: papers.ssrn.com/sol3/papers....

The results? LLM-generated memorials often scored average to superior, some even receiving exceptional praise! 🌟 However, our analysis also uncovered persistent shortcomings: factual inaccuracies, hallucinated citations, and superficial legal analysis. 🧐

We put leading LLMs (Gemini 2.0 & GPT-4o) to the ultimate test: crafting complete memorials for the prestigious Jessup Moot Court Competition. These AI-generated submissions were then anonymously evaluated by real judges – a unique benchmark!

Thrilled to share our new paper, "GenAI as an International Lawyer: A Case Study with the Jessup International Law Moot Court", which Damien Charlotin and myself have been working on! ⚖️🤖

SSRN Link: papers.ssrn.com/sol3/papers....

Digital Futures Institute announces eleven new Fellows for 2025-26 Eleven interdisciplinary projects awarded Fellowships from the Digital Futures Institute for the 2025-26 cohort.

👏 Meet our eleven new Digital Futures Institute Fellows!

Congratulations @lboungr.bsky.social, Rowan Boyson, Mark Cote, Amrita Dhillon, Alex Gould, Elisabeth Kelan, @niccoloridi.bsky.social, Gabriele Salciute Civiliene, Astrid Van den Bossche, @jamiewoodcock.bsky.social & @lorenzoz.bsky.social 🔽

From Art. 38(1)(d) ICJ Statute to LLMs: how will AI reshape "the teachings of the most highly qualified publicists"? On textbooks, authority, and knowledge production in international law in the age of artificial intelligence.

papers.ssrn.com/sol3/papers....

Congrats Jim!

Welcome to day 2 of me ignoring all my other commitments to watch the ICJ climate advisory opinion live stream. Day 2 will feature interventions by Belize, Bolivia, Brazil, Burkina Faso, Cameroon, the Philippines, Canada, Chile, China, Colombia, Dominica, and South Korea webtv.un.org/en/asset/k1p...

The afterlife of ISDS awards: post-award settlements and the limits of transparency reforms Abstract. Post-award settlements in investor–state dispute settlement (ISDS) remain largely unstudied. One of the reasons for this gap in knowledge is the

Finally, it’s out! Our paper paper on the hidden afterlife of ISDS awards published in open access: academic.oup.com/jids/advance...
We draw on a new COPIID dataset to unveil patters of compliance with investment arbitration awards and unveil a significant number of post-award settlements…

Mapping the Invisible College of IL + the neighbouring law professoriate more generally. An interactive visualization of a "friends of friends" network (with some pruning, lest it become unwieldy). Takes a while to load, will likely won't open on mobile.

ouestware.gitlab.io/retina/beta/...

The idea is to focus on likes, replies, and reposts to identify patterns of engagement—who supports, who debates, and whose ideas are amplified.

I start with one user and expands outward, capturing connections to followers, those they follow, and interactions across posts. So I can identify both direct and indirect ties within these networks.

I’ve been analysing how international lawyers (and their 'neighbours') interact on Bluesky, creating networks based on likes, replies, reposts, and connections.

Thanks! There will be more as soon as the numbers are crunched!

International legal argument / banter unfolding: we have a proof of concept.

Total nodes collected: 141112
Total interactions: 403857

- Likes: 186693 total interactions
- Replies: 108617 total interactions
- Reposts: 108547 total interactions

Still calculating everything—it got to 1.5M edges pretty quick, so I will need to implement some culling logic to make it more manageable!

1) Unpacking international legal argument by leveraging computational techniques and, more recently, GenAI;
2) Mapping the 'invisible college' of international lawyers and its dynamics in various ways, including scraping Bluesky. (Not technically working on it, but the code is running...)

Okay, I've found most of you. Hello Invisible College of BlueSky!

Welp—and it's just 10% of the data.

Hold my beer, I am going in!

I guess we'll know soon: