RAG Automation Platform

A production-oriented retrieval-augmented generation project built with Java, Spring Boot, n8n, PostgreSQL/pgvector, MongoDB, and Redis. It runs without an API key in local mode and can use a local Ollama model for generated, citation-aware answers.

What it does

• Ingests text through the REST API, dashboard, or an n8n webhook.
• Splits documents into overlapping chunks and creates deterministic 384-dimensional embeddings.
• Retrieves the closest chunks in memory for a zero-setup demo or from PostgreSQL with pgvector in production.
• Produces keyless extractive answers, with optional Ollama generation.
• Caches repeat questions in Redis and invalidates them when the knowledge base changes.
• Stores query and ingestion audit events in MongoDB.
• Exposes health, metrics, and service status endpoints.
• Builds and tests every change with GitHub Actions, then publishes the application image to GHCR.

Architecture

flowchart LR
    User[Dashboard or API client] --> API[Spring Boot RAG API]
    Automation[n8n webhook] --> API
    API --> Chunk[Chunk and embed]
    Chunk --> PG[(PostgreSQL + pgvector)]
    API <--> Redis[(Redis query cache)]
    API --> Mongo[(MongoDB audit log)]
    API -. optional .-> Ollama[Ollama chat model]
    Actions[GitHub Actions] --> Tests[Tests and image build]
    Actions --> GHCR[GitHub Container Registry]

Loading

Component	Responsibility
Spring Boot	REST API, validation, retrieval, answer orchestration, dashboard
PostgreSQL + pgvector	Documents, chunks, vector similarity search
MongoDB	Append-only ingestion and query audit events
Redis	Time-limited query response cache
n8n	Validated document-ingestion webhook automation
Ollama	Optional local answer generation; no cloud key required

Run locally now

Local mode needs only Java 21. It keeps vectors, cache entries, and audit events in memory.

.\mvnw.cmd spring-boot:run

Open http://localhost:8080, or run the automated smoke test in a second PowerShell window:

.\scripts\smoke-test.ps1

Run the complete stack

Install Docker Desktop, then:

Copy-Item .env.example .env
# Change the example passwords and encryption key in .env.
docker compose up --build

The services are:

• Dashboard and API: http://localhost:8080
• n8n: http://localhost:5678
• Spring health: http://localhost:8080/actuator/health

The compose stack imports and activates RAG - Document Ingestion Webhook. After creating the first n8n owner account, call its production webhook:

$body = @{
  title = "Support handbook"
  source = "n8n-webhook"
  content = "Priority one incidents require an acknowledgement within fifteen minutes."
} | ConvertTo-Json

Invoke-RestMethod `
  -Method Post `
  -Uri http://localhost:5678/webhook/knowledge-ingest `
  -ContentType application/json `
  -Body $body

Optional Ollama answers

The default answer provider quotes the best matching passages and needs no model download. To generate a composed answer, install Ollama, pull a model, and set this in .env:

ollama pull llama3.2:3b

ANSWER_PROVIDER=ollama
OLLAMA_BASE_URL=http://host.docker.internal:11434
OLLAMA_CHAT_MODEL=llama3.2:3b

Restart the application service after changing the values:

docker compose up -d --build app

API

Ingest a document

POST /api/v1/documents

{
  "title": "Incident handbook",
  "source": "internal-runbook",
  "content": "When Redis is unavailable, restart the service and verify connectivity."
}

Ask a question

POST /api/v1/query

{
  "question": "How should Redis be recovered?",
  "topK": 4,
  "conversationId": "incident-42"
}

The response contains the answer, similarity-ranked source chunks, their scores, a cache flag, and a conversation ID.

Inspect runtime status

GET /api/v1/status

The endpoint reports the document count and active vector, cache, audit, and answer providers.

Configuration

Variable	Default	Purpose
SPRING_PROFILES_ACTIVE	default local profile	Use production for the database-backed stack
ANSWER_PROVIDER	extractive	Set to ollama for generated answers
OLLAMA_BASE_URL	http://localhost:11434	Ollama API address
OLLAMA_CHAT_MODEL	llama3.2:3b	Ollama chat model
POSTGRES_*	development values	PostgreSQL connection and initialization
MONGODB_URI	local development URI	MongoDB audit database
REDIS_PASSWORD	ragredis	Redis authentication

See .env.example for the complete deployment template.

Development

Run all tests and create the executable JAR:

.\mvnw.cmd verify

The project includes tests for chunk overlap, normalized embeddings, ingestion, retrieval, cache behavior, API status, and validation. CI also validates the Compose file and builds the container image.

Deployment

Every push to main runs CI and publishes:

ghcr.io/javaprogswing/rag-n8n-spring-platform:latest

Pull that image into any Docker or Kubernetes environment and provide the same production environment variables shown in compose.yml.

License

Released under the MIT License.