Building an Efficient RAG Knowledge-Based Q&A System with Spring Boot and openGauss DataVec ​

@Configuration
public class opgsConfig {
    @Value("${spring.datasource.url}")
    private String url;

    @Value("${spring.datasource.username}")
    private String username;

    @Value("${spring.datasource.password}")
    private String password;

    @Value("${spring.datasource.driver-class-name}")
    private String driver;

    public Connection getConnection() {
        // Connect to the database
    }
}

@Repository
public class Repository {
    private Connection conn;

    public void CreateTable(int dim)
    {
        ...
    }

    public void CreateIndex()
    {
        ...
    }

    public void InsertDataSingle(int id, String content, String vector)
    {
        ...
    }

    public String findNearestVectors(int efsearch, String vector, int topK)
    {
        ...
    }
    ...
}

@Service
public class Service {
    private final Repository repository;

    @Value("${ollama.modelDim}")
    private int vectorDim;

    @Value("${ollama.embeddingURL}")
    private String embeddingURL;

    @Value("${ollama.model}")
    private String ollamaModel;


    // Call the Ollama embedding service
    public float[] getEmbedding(String message)
    {
        OllamaApi ollamaApi = new OllamaApi(embeddingURL);
        OllamaOptions options = OllamaOptions.builder().withModel(ollamaModel).build();
        OllamaEmbeddingModel embeddingModel = new OllamaEmbeddingModel(ollamaApi, options);
        EmbeddingResponse embeddingResponse = embeddingModel.call(new EmbeddingRequest(List.of(message), options));
        return embeddingResponse.getResult().getOutput();
    }

    // Call the APIs for interacting with the vector database
    public void CreateTxtTable()
    {
        repository.CreateTable(vectorDim);
    }

    public void InsertTuples(int id, String message)
    {
        float[] res = getEmbedding(message);
        repository.InsertDataSingle(id, message, Arrays.toString(res));
    }

    public void IndexTxt()
    {
        repository.CreateIndex();
    }

    public String QueryContent(int efsearch, String query, int topK)
    {
        float[] res = getEmbedding(query);
        return repository.findNearestVectors(efsearch, Arrays.toString(res), int topK);
    }

    public String BuildRagPrompt(String query, List<String> relatedContents, int maxLength)
    {
        List<String> seletedTexts = new ArrayList<>();
        int totalLen = 0;
        for (String content : relatedContents) {
            if (totalLen + content.length() < maxLength) {
                seletedTexts.add(content);
            } else {
                break;
            }
        }
        return String.format("Answer the following question concisely in English based on the information below:\n%s\nQuestion: %s\nAnswer:",
                             String.join("\n", seletedTexts),
                             query);
    }
    ...
}

@RestController
public class Controller {
    @Autowired
    private Service service;

    private final ChatModel chatModel;

    public Controller(ChatModel chatModel)
    {
        this.chatModel = chatModel;
    }

    @GetMapping("/index")
    public String IndexDoc()
    {
        service.CreateTxtTable();
        service.InsertTuples(0, "Large-scale pre-trained language models support efficient parallel training and multiple NLP tasks.");
        service.InsertTuples(1, "Multimodal fusion models combine text, image, and audio inputs to provide comprehensive data understanding.");
        service.InsertTuples(2, "Distributed deep learning frameworks are easy to scale and support large-scale data processing.");
        service.InsertTuples(3, "Video understanding and generation models use advanced time-series analysis for monitoring and entertainment applications.");
        service.InsertTuples(4, "Ultra-high-resolution image generation models based on GAN architecture capture fine details effectively.");
        service.IndexTxt();
        return "embedding and index succeed!";

    }

    @GetMapping(value = "/chat", produces = "text/plain;charset=UTF-8")
    public String queryVector(@RequestParam(value = "message", defaultValue = "Briefly introduce openGauss.") String query)
    {
        // String query = "What are the large language models suitable for efficient parallel training?";
        int topK = 2;
        int maxPromptLen = 100;

        List<String> res = service.QueryContent(2, query, topK);

        String generatePrompt = service.BuildRagPrompt(query, res, maxPromptLen);
        System.out.println("prompt:");
        System.out.println(generatePrompt);

        Flux<ChatResponse> stream = chatModel.stream(new Prompt(generatePrompt));
        return stream.map(resp -> resp.getResult().getOutput().getText());
    }

}