What it is
A local RAG Q&A system for the Chinese server's League of Legends patch notes. Numbers in answers come only from structured extraction of official announcements, never from a generative model. Built for users asking about champion, item, and system changes across 30 versions.
How it uses Jev
Jev takes retrieval candidates plus the parsed query state and picks which candidate to use (reranking), and verifies each answer against its evidence (self-check). A confidence gate skips reordering when noul score gap is below 0.15. Without an API key, the system degrades gracefully and records this in the trace.
Primitives:choicenoul
Technique worth stealing
Metadata filtering before ranking, so candidates shrink to 1-6 and reranking only resolves genuine ambiguity.
Try it
Set TYPESAFE_API_KEY, then run python launch.py and open http://127.0.0.1:18890
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.67
- Shows a System One patternyes0.90
- Handles uncertaintyunclear0.41
- Measuredyes0.93
- Runnableunclear0.49
- Worth recommendingyes0.61
- Model replicaunclear0.44
- Problem scopescore on a 0–2 scale0.85
- About Jevyes0.96
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.