Spring AI Showcase

A simple project demonstrating Retrieval-Augmented Generation (RAG) using:

• Spring Boot 4.0.6 / Java 23
• Spring AI 2.0.0-M6
• PGVector as the vector database
• OpenAI for embeddings (text-embedding-ada-002) and chat (gpt-4o-mini)

---

How RAG works in this project

              ┌─────────────────── STARTUP (once) ───────────────────────┐
              │  TIOBE.pdf  ──►  PagePdfDocumentReader                   │
              │                        │                                  │
              │                 TokenTextSplitter (chunks)               │
              │                        │                                  │
              │              OpenAI Embeddings API                       │
              │                        │                                  │
              │                   PGVector DB  ◄────────────────────────  │
              └───────────────────────────────────────────────────────── ┘

              ┌─────────────────── GET /describe ────────────────────────┐
              │  Query ──► Embedding ──► Similarity Search (PGVector)    │
              │                              │                            │
              │                     Top-5 chunks (context)              │
              │                              │                            │
              │                    OpenAI Chat API (gpt-4o-mini)        │
              │                              │                            │
              │                         Response (JSON)                  │
              └───────────────────────────────────────────────────────── ┘

---

Setup

1. Add TIOBE.pdf

Place the PDF file at:

src/main/resources/TIOBE.pdf

2. Set your OpenAI API key

cp .env.example .env
# Edit .env and insert your key: OPENAI_API_KEY=sk-...

3. Start with Docker Compose

docker compose up --build

4. Call the endpoint

curl http://localhost:8081/describe

---

Console output

On startup:

>>> [APP]     Starting Application...
>>> [INGEST]  Step 1/4 - Loading 'TIOBE.pdf' from classpath...
>>> [INGEST]  Step 2/4 - Reading PDF pages with PagePdfDocumentReader...
>>> [INGEST]  Step 3/4 - Splitting pages into token chunks...
>>> [INGEST]  Step 4/4 - Computing embeddings and storing in PGVector...

On /describe:

>>> [CONTROLLER]  GET /describe received.
>>> [RAG]         Step 1/3 - Searching relevant chunks in VectorStore (TopK=5)...
>>> [RAG]         Step 2/3 - Building context from chunks...
>>> [RAG]         Step 3/3 - Sending request with context to OpenAI LLM...

---

Note on duplicates

The document is re-ingested into PGVector on every app start. For production use, you should check whether the document has already been loaded (e.g. via a metadata check or a separate initialization flag table).