langchain4j-vector-stores-configuration

The langchain4j-vector-stores-configuration skill provides configuration patterns for LangChain4j vector stores used in Retrieval-Augmented Generation applications, solving the problem of setting up embedding storage and retrieval consistently across multiple database backends. It offers a unified abstraction over PostgreSQL/pgvector, Pinecone, MongoDB Atlas, Milvus, and Neo4j with builder-based configuration, metadata filtering, and hybrid search.

The skill shows how to set up a basic embedding store (for example, a PgVectorEmbeddingStore with host, port, database, table, dimension, and indexing options), and prescribes a validation workflow: build the store, run a connectivity health check, confirm embedding dimensions match, ingest a small test batch, run a sample semantic search to verify retrieval, and only then proceed to full production ingestion. It also demonstrates configuring multiple qualified stores for different use cases (such as separate document and chat-history stores), and automated document ingestion with EmbeddingStoreIngestor using a recursive document splitter and a tokenizer. It bundles a SKILL.md, an examples reference, and an API reference, and declares safe allowed-tools (Read, Write, Edit, Bash, Glob, Grep). The examples include placeholder database credentials as standard connection configuration, not exfiltration.

It targets Java backend and AI engineers building semantic search and production RAG systems who need to integrate vector databases, set up metadata filtering and hybrid search, and optimize vector store performance. Use cases include document embedding pipelines, multi-store architectures, and dimension validation before production rollout.