Docs
OfficialJev Cookbooks
Cookbook: Knowledge Graph Entity Alignment
Decide which of 450 candidate product pairs from two beer catalogues describe the same entity: one three-level Score per pair splits 40 merges, 50 curator reviews, and 360 non-matches.
- Format
- Docs
- Reading time
- 3 min
- Published by
- TypeSafe AI
- Last verified
- Sep 2026
Highlights
- 450 Magellan beer pairs scored in one request each: 40 assert sameAs (8.9%), 50 go to a curator (11.1%), and 360 stay unlinked.
- No fitted threshold: route() rounds the Score to the nearest of three levels, so the middle level's wording drives behavior.
- Three Noul questions about name, brewery, and style ride along to localize disagreements, while ABV stays in arithmetic.
Quickstart
python
response = client.system_one(
state={"entity_a": pair["entity_a"], "entity_b": pair["entity_b"]},
questions={
"link_state": Score(instructions="How do the two products relate?",
criteria=LEVELS), # different / related / same
"same_name": Noul(instructions="Same beer name?"),
"same_brewery": Noul(instructions="Same brewery?"),
"same_style": Noul(instructions="Same beer style?"),
},
model="jev-1.12",
)
score = response.answers["link_state"].score
outcome = OUTCOME[min(int(score + 0.5), len(LEVELS) - 1)] # nearest level routes the pairWatch out
A rounded three-level score is a routing policy rather than a calibrated match probability, so keep a curator path for the middle.
Questions about this entry
Short answers drawn from the linked project or page.
The cookbook routes by rounding the Score to the nearest of three levels, so the middle level's wording is the policy. Of 450 beer pairs, 40 assert sameAs, 50 go to a curator, and 360 stay unlinked.
Localization. Three Nouls about name, brewery, and style ride along to show where a disagreement comes from, while ABV comparisons stay in arithmetic rather than the model.
Because merging two entities inappropriately is the more expensive mistake: every fact about either entity then describes the merged one, and anything linked to either comes along too, with undo meaning working out which fact came from where. Missing a match only leaves a duplicate. So the judgment needs an option for pairs that are neither safe to merge nor safe to drop.
One level per outcome: different product, leave the two entities unlinked; related but possibly not the same, hand it to a curator; same product, merge them. Score is used so the semantic label attaches directly to each outcome including the middle one. A Noul could do it indirectly by thresholding its output, and a Choice would lose the ordering between the three.
The Beer data from the Magellan collection: two beer catalogues scraped from different websites, already cut to 450 candidate pairs by a cheap rough first pass. Each entity carries name, brewery, style and alcohol content, and the text is left exactly as published, with unconverted HTML entities, apostrophes split off as separate words and a few characters decoded wrongly, so no pre-processing flatters the model.
pip install matplotlib ipython 'cooksafe>=0.2.0,<0.3.0' and set TYPESAFE_API_KEY; the published numbers came from jev-1.12 on 2026-08-11. One request goes out per pair, so spend follows the number of pairs you were handed rather than the size of either source. The example uses a small pool of 6 workers because the public endpoint rate-limits above roughly eight.
No. The caveat describes a rounded three-level score as a routing policy rather than a calibrated match probability, which is why the middle level keeps a curator path. The three companion Nouls only localize which field disagrees, and ABV comparisons stay in arithmetic rather than the model.
More Jev resources like this
A 14-Noul claims-triage rubric over one insurance claim, repeated 15 times: mean probability standard deviation of 0.0102 at 111 ms per call, with a 0.30 to 0.70 band for human review.
Jev CookbooksDocs#official#cookbook#consistency
Official
A two-stage extraction cascade: a mini model extracts, a Noul battery verifies each field in one request, and a 0.7 gate escalates to a reasoning model, sitting on the cost/quality frontier.
Jev CookbooksDocs#official#cookbook#extraction
Official
Score retrieved passages with four Noul questions each, then route in code: prompt injections dropped, false premises kept as conflicts, and evidence assembled separately for the generator.
Jev CookbooksDocs#official#cookbook#search
Official
Turn wine tasting notes into 67 numeric columns for a CatBoost regressor: five rounds of reading its own errors cut held-out RMSE from 3.09 to 1.77 across 800 unseen reviews.
Jev CookbooksDocs#official#cookbook#research
Official
Fifteen runnable OpenRouter recipes for Jev covering support triage, database indexing, PII scanning, moderation, lead scoring, a browser agent, and a Gmail labeler, each with a small labelled dataset and saved live results.
Jev Cookbooks#community#javascript#openrouter
Community
nexibeo/jev-cookbookJavaScript
nexibeo/jev-cookbookJavaScriptA support-ticket cookbook that lets Jev decide which requests are worth remembering, then wraps each call so saved tickets ride along in the state. Six later messages route correctly with memory against two without.
Jev CookbooksDocs#official#cookbook#memory
Official
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.