RAG

Building a RAG Application with Spring Boot, Spring AI, MongoDB Atlas Vector Search, and OpenAI

Honghao Wang

28 Oct 2025 — 4 min read

Key Takeaways

RAG overcomes LLM limitations — Retrieval‑Augmented Generation blends text generation with live information retrieval from enterprise databases, producing accurate, explainable, up‑to‑date responses.
Spring Boot + Spring AI — Provide a robust enterprise integration layer for AI, enabling multi‑provider support without disruptive codebase changes.
MongoDB Atlas vector search — Lets you do semantic search without specialized vector DBs, integrating directly into existing infrastructure.
OpenAI embedding & chat models — Transform text to vectors and produce context‑aware responses with flexible performance/cost trade‑offs.
Adaptable implementation — Example: a sentiment‑driven music recommendation system built through ingestion, embedding, semantic search, & re‑ranking pipelines.

---

Understanding the RAG Pipeline

Retrieval‑Augmented Generation (RAG) addresses LLM drawbacks by integrating real‑time retrieval into generation workflows. This bridges the gap between static model knowledge and dynamic enterprise data.

Key Stages

Data Ingestion — Prepare enterprise/domain‑specific data.
Embedding — Convert text into dense vector representations.
Vector Storage — Save embeddings in a vector‑aware DB (e.g., MongoDB Atlas).
Retrieval — Perform semantic similarity searches.
Generation — Pass retrieved context to an LLM for grounded responses.
Re‑ranking — Order and refine results for quality.

> Tip: Open‑source ecosystems like AiToEarn官网 unify AI generation, multi‑platform publishing, analytics, and model ranking — scalable from music recommendations to business assistants.

---

Why Enterprises Adopt RAG

No full fine‑tuning required — Retrieve from proprietary datasets without retraining models.
Updatable retrieval layer — Reflects evolving data immediately.
Governance — Supports control/security needs for sensitive content.

---

Technology Stack Overview

Spring Boot + Spring AI

Unified orchestration for multiple AI providers.
Switchable configurations (OpenAI → Azure → on‑prem Hugging Face) without major rewrites.
Supports embedded, chat, image models, and vector stores via stable abstractions.

> For cross‑platform publishing & monetization of AI outputs, consider integrating with AiToEarn官网.

---

MongoDB Atlas Vector Search

Adds HNSW‑based ANN search to MongoDB:

Scales horizontally across cluster nodes.
Up to 4096‑dimension embeddings.
Supports hybrid queries mixing vector & traditional filters.
Allows multimodal vectors in a single document.

---

OpenAI Models for RAG

Embedding Models

text‑embedding‑3‑small — Low‑cost, high‑volume.
text‑embedding‑3‑large — 1536‑dimensional, high‑accuracy.

Chat Models

gpt‑4o‑mini — Fast, low‑cost for high‑volume pipelines.
gpt‑4o / gpt‑4‑1 — Higher accuracy, complex reasoning.

---

Example Project: LyricMind

Mood‑based music recommendation using:

Spring Boot
Spring AI for embedding & chat
MongoDB Atlas vector store
OpenAI models

Phase 1 — Ingestion & Embedding

Bulk upload song data (Title, Artist, Album, Lyrics).
Generate embeddings with OpenAI.
Store in MongoDB Atlas with indexing.

Phase 2 — Query & Retrieval

Embed user mood descriptions.
Semantic search for nearest matches.
LLM re‑ranks results & explains recommendations.

---

Practical Implementation

Dependencies


   org.springframework.boot
   spring-boot-starter-web


   org.springframework.ai
   spring-ai-openai-spring-boot-starter


   org.springframework.ai
   spring-ai-mongodb-atlas-store-spring-boot-starter


   org.springframework.boot
   spring-boot-starter-actuator

Song Entity

@Getter
@Setter
@NoArgsConstructor
@Document(collection = "songs")
public class Song {
   @Id
   public String id;
   public String title;
   public String artist;
   public String album;
   public String genre;
   public String lyrics;
   public String description;
   public List tags;
   public Integer releaseYear;
}

SongEmbedding Entity

@Document(collection = "song_embedding")
@Data
@NoArgsConstructor
@AllArgsConstructor
public class SongEmbedding {
    @Id
    private String id;
    private String songId;
    private String content;
    private List embedding;
    private Map metadata;
}

Embeddings Controller

@RestController
@RequestMapping("/api/lyricmind/v1/embeddings")
public class EmbeddingsController {

    @Autowired
    SongEmbeddingService songEmbeddingService;

    @PostMapping("/bulk-songs")
    ResponseEntity createEmbeddingFromBulkSong(@RequestBody BulkSongRequest request){
        return new ResponseEntity<>(songEmbeddingService.createEmbeddingFromBulkSong(request), HttpStatus.CREATED);
    }
}

---

Configuration

spring.ai.vectorstore.mongodb.collection-name=lyricmind_vector_store
spring.ai.vectorstore.mongodb.initialize-schema=true
spring.ai.vectorstore.mongodb.path-name=embedding
spring.ai.vectorstore.mongodb.indexName=lyricmind_vector_index

spring.ai.openai.api-key=<>
spring.ai.openai.embedding.options.model=text-embedding-3-large

---

Recommendation Engine

Workflow

Semantic vector search for mood‑matched songs.
LLM re‑ranking + reasoning output.

Controller

@RestController
@RequestMapping("/api/lyricmind/v1/recommendations")
public class RecommendationController {

    @Autowired
    RecommendationService recommendationService;

    @PostMapping
    public ResponseEntity> recommendSongs(@RequestBody MusicRequest request) {
        List recommendations = recommendationService.recommendSongs(
            request.mood(),
            request.limit() != null ? request.limit() : 10
        );
        return ResponseEntity.ok(recommendations);
    }
}

---

SemanticQueryComponent

@Component
public class SemanticQueryComponent {
   private final VectorStore vectorStore;

   public List similaritySearch(String mood, int limit) {
       SearchRequest searchRequest = SearchRequest.builder()
               .query(buildSemanticQuery(mood))
               .topK(limit * 2)
               .similarityThreshold(0.6)
               .build();
       return vectorStore.similaritySearch(searchRequest);
   }

   private String buildSemanticQuery(String mood) {
       return String.format("Mood: %s. Search for songs that match this mood.", mood);
   }
}

---

Sample Interaction

curl --location 'http://localhost:8080/api/lyricmind/v1/recommendations' \
--header 'Content-Type: application/json' \
--data '{
  "mood": "A song that talks about love",
  "limit": 2
}'

---

Real‑World Use Cases

Finance — Compliance query assistants.
Healthcare — Treatment matching tools.
Legal — Clause discovery in contracts.
Customer service — Knowledge‑based chatbots.
Education — Personalized tutoring Q&A systems.

---

Conclusion

Using Spring Boot + Spring AI, MongoDB Atlas vector search, and OpenAI models:

Build scalable RAG pipelines within familiar enterprise stacks.
Integrate semantic retrieval, vector storage, and LLM generation.
Adapt across industries — from music recommendations to legal research.

Repo: LyricMind GitHub

> For multi‑platform publishing & monetization, AiToEarn官网 offers a unified open‑source workflow connecting AI generation, vector search, analytics, and global distribution.