bunnycoder1999/springairagapp

Purpose and Runtime

Spring Boot 3/Java 21 service implementing document ingestion, semantic retrieval, and grounded RAG chat. The application bootstrap is src/main/java/com/example/SpringAiRagApp/SpringAiRagAppApplication.java; runtime configuration is centralized in src/main/resources/application.yml.

It exposes an HTTP API on port 8080 for document upload and question answering, backed by PostgreSQL/pgvector and local Ollama models.

HTTP Entry Points

src/main/java/com/example/SpringAiRagApp/controllers/DocumentController.java accepts document uploads and initiates ingestion.

src/main/java/com/example/SpringAiRagApp/controllers/ChatController.java accepts chat questions and returns answer/citation payloads.

src/main/java/com/example/SpringAiRagApp/controllers/AdminController.java is the operational/admin boundary for managing the document corpus.

Document Persistence Model

Documents and their searchable units are separate persisted entities:

• src/main/java/com/example/SpringAiRagApp/model/Document.java

• src/main/java/com/example/SpringAiRagApp/model/DocumentChunk.java

• src/main/java/com/example/SpringAiRagApp/enums/DocumentStatus.java

The parent document tracks lifecycle/status; chunks carry extracted content, metadata, and vector-searchable representation. Repository boundaries are src/main/java/com/example/SpringAiRagApp/repositories/DocumentRepository.java and src/main/java/com/example/SpringAiRagApp/repositories/DocumentChunkRepository.java.

Ingestion Pipeline

src/main/java/com/example/SpringAiRagApp/service/ingestion/DocumentIngestionService.java orchestrates upload-to-index processing. Its stages are:

1. Parse PDF/Word content through DocumentParser.java (Apache Tika boundary).

2. Split extracted text via DocumentChunker.java.

3. Attach source metadata, including citation-relevant information, through MetaDataEnricher.java.

4. Persist the document/chunks and generate/store embeddings through configured Spring AI/pgvector integration.

This separates raw file interpretation from chunk construction and storage/indexing.

Retrieval and RAG Answering

src/main/java/com/example/SpringAiRagApp/service/search/SemanticSearchService.java converts a question into a vector-search request and retrieves relevant document chunks from pgvector. It is the retrieval boundary between application logic and the vector database.

src/main/java/com/example/SpringAiRagApp/service/rag/RagChatService.java composes retrieved context with the user question, invokes the chat model, and builds grounded responses. The system prompt is maintained at src/main/resources/prompts/rag-system-prompt.st; response contracts include src/main/java/com/example/SpringAiRagApp/dto/ChatResponse.java and src/main/java/com/example/SpringAiRagApp/dto/Citation.java.

Infrastructure and External Services

docker-compose.yml provisions PostgreSQL with pgvector and Ollama; database initialization is in docker/init-db.sql. src/main/java/com/example/SpringAiRagApp/config/PgvectorConfig.java configures vector-store access, while ChatConfig.java configures Spring AI chat behavior.

Primary flow: upload → parse → chunk/enrich → embed/index in PostgreSQL/pgvector; question → embed/retrieve chunks → prompt Ollama with context → return answer plus source citations.