Article
CommunityJev Integrations & Adapters
Spring AI Modular RAG with Jev
Christian Tzolov plugs JevDocumentFilter and JevDocumentReranker into Spring AI's Modular RAG pipeline: four Noul questions per chunk drop injections, contradictions, and noise before a one-question reranker keeps the top three.
- Format
- Article
- Reading time
- 2 min
- Published by
- Community
- Author
- Christian Tzolov
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Java
- Capabilities
- Open sourceRequires API key
Highlights
- Filter defaults: prompt injection excludes above 0.70, contradictions stay tagged above 0.70, relevance fails below 0.45, and answer evidence passes above 0.55.
- JevDocumentReranker asks one question, whether a passage could answer the query, sorts the survivors by score, and keeps the top K under jev.rerank.score.
- The live run expanded one chatty question into four searches; six chunks passed similarity 0.5, and only two survived at 0.97 and 0.88.
- Filter first, rerank second, so the reranker pays only for survivors; contradicting passages survive on purpose so a false premise meets counter-evidence.
- The demo needs spring-ai-rag plus the TypeSafe 0.3.0 artifacts, and a TaskExecutor must be passed or non-daemon threads keep the app alive.
Watch out
Vendor-authored by the project author and demo-focused: the thresholds live in a Policy record you can override and no accuracy benchmarks are given. Both components send one Jev call per passage, so cost scales with chunk count, and newer Claude models reject temperature, which the demo works around by cloning the builder.
Questions about this entry
Short answers drawn from the linked project or page.
JevDocumentFilter and JevDocumentReranker are DocumentPostProcessors in Spring AI's Modular RAG pipeline: the filter drops prompt injections and passages without answer evidence, and the reranker scores the survivors and keeps the top K.
Add spring-ai-rag from the Spring AI BOM plus spring-ai-starter-typesafe and typesafe-spring-ai at 0.3.0, set spring.ai.typesafe.api-key, ingest with PagePdfDocumentReader and TokenTextSplitter, and the starter auto-configures TypeSafeClient.
Four Noul questions per passage: contains_prompt_injection excludes above 0.70, contradicts_query_premise keeps above 0.70 tagged CONFLICTING, is_relevant excludes below 0.45, and contains_answer_evidence keeps above 0.55. The result is stored under jev.classification.
Order matters: the filter runs first so the reranker only pays for passages that passed, then it asks a single question, could this passage answer the query, scores each survivor, and keeps the top K under jev.rerank.score.
On purpose: if the user assumes something false, the model should see the evidence that says so. Passages contradicting the query premise are kept and tagged CONFLICTING rather than excluded.
The rewrite and expansion turned one chatty sentence into four focused searches; the joiner produced six unique chunks above similarity 0.5, Jev kept two at scores 0.97 and 0.88, and topK(3) returned only those two, so the model got less context, all of it relevant.
The default executor keeps a command-line app alive on non-daemon threads, so pass Spring Boot's TaskExecutor via .taskExecutor(...) or enable virtual threads. Newer Claude models reject temperature, so the demo clones the builder so the main client's advisors stay out of rewrite and expand calls.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: A RAG loop where Jev decides what is real
X: Measuring what Jev does in a RAG pipeline
X: Jev-powered reranking for RAG. Likitd_’s jev-ranker uses Jev to rerank results in a RAG pipeline.
X: Coverage is not product value
More Jev resources like this
Spring AI Community's TypeSafe starter puts Jev behind seven Spring AI SPIs: judge, self-refine advisor, guardrail advisor, document filter, reranker, tool index, and evaluator.
Jev Integrations & AdaptersArticle#community#java#spring
Community
Graph RAG framework over Milvus vector search with no graph database required: Jev Noul judgments select graph relations and rerank full source passages alongside vector-search candidates, reaching 86.07% average Recall@5 on 1,000 MuSiQue and 1,000 2Wiki queries in the two-stage evaluation.
Jev Integrations & Adapters#community#rag#retrieval
Community
zilliztech/vector-graph-ragPython
zilliztech/vector-graph-ragPythonLocal-first knowledge search with seven selectable pipelines: BM25 plus Jev reranking is the default vector-free path, plus agentic, embedding-hybrid, taxonomy, gate and line-search modes. Reranks up to 30 candidates in one batched set of Jev Noul relevance questions and streams cited answers from a configurable model.
Jev Integrations & Adapters#community#rag#retrieval
Community
aifabrice/jev-ragPython
aifabrice/jev-ragPythonVolcengine's agent context database that unifies memory, knowledge RAG and skills under a viking:// virtual filesystem; it ships a Jev rerank client that scores candidate documents against api.typesafe.ai and treats the probability as relevance.
Jev Integrations & Adapters#community#agent-memory#rag
Community
volcengine/OpenVikingPython
volcengine/OpenVikingPythonPrecision-first RAG orchestrator that wires Jev as a System 1 gate around an LLM: it routes chitchat, triages retrieved chunks with relevance and answer-evidence Nouls, then verifies each claim and withholds drafts whose claims are unsupported.
Jev Integrations & Adapters#community#rag#hallucination
Community
david96182/cribrixPython
david96182/cribrixPythonLong-document search that pairs Jev's Choice primitive with PageIndex's hierarchical tree: instead of a vector DB or embeddings, Jev picks a node and then a child, so even a 300-page report can be searched a handful of options at a time.
Jev Integrations & Adapters#community#github#rag
Community
VectifyAI/jev-doc-searchPython
VectifyAI/jev-doc-searchPythonFrom 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.