Stuck on implementing RAG agent with OpenAI

[TabestoJulienDuvignau]

Hello,

I'm trying to implement a RAG chatbot, but I'm stuck on its implementation and I don't know what I've done wrong.

I created a Symfony command to upload my files. The code is very simple:

RAGChatBotAgent::make($this->openaiApiKey, $this->logger, $this->inspector) 
    ->addDocuments(FileDataLoader::for(__DIR__.'/../../documents/history.md')->getDocuments());

When I run the command, I get no errors and a neuron.store file is created with the contents of my history.md file.
When I go to the OpenAI interface, no VectorStore has been created.

Then my agent looks like this:

class RAGChatBotAgent extends RAG
{
    protected string $apiKey;
    protected LoggerInterface $logger;
    
    public function __construct(string $apiKey, LoggerInterface $logger)
    {
        $this->apiKey = $apiKey;
        $this->logger = $logger;
    }

    protected function provider(): AIProviderInterface
    {
        return new OpenAI(
            key: $this->apiKey,
            model: 'gpt-3.5-turbo',
        );
    }
    
    protected function embeddings(): EmbeddingsProviderInterface
    {
        return new OpenAIEmbeddingsProvider(
            key: $this->apiKey,
            model: 'text-embedding-3-small'
        );
    }
    
    protected function vectorStore(): VectorStoreInterface
    {
        return new FileVectorStore(
            directory: __DIR__ . '/../../neuronStore'
        );
    }
    
    public function instructions(): string
    {
        $this->logger->debug('RAGChatBotAgent: Setting up system instructions');
        
        return new SystemPrompt(
            background: ["You are a helpful AI assistant specialized in answering questions about past technical incidents."]
        );
    }
}

And of course, when I instantiate the agent to ask it a question, it replies that it does not know how to answer me:

            // Create a user message with the question
            $userMessage = new UserMessage($question);
            
            // Use the chat method from the Agent class with the user message and monitoring
            $response = $chatbotAgent
                ->observe(new AgentMonitoring($this->inspector))
                ->answer($userMessage);

Does anyone know how to help me solve this problem? What did i miss?
Thank you very much!

[ilvalerione]

Hi @TabestoJulienDuvignau you are using OpenAI only to calculate embeddings not as a vector store. You are using the FileVectorStore to store embedded documents in your file system.

Are you getting some errors or was this your only doubt?

[TabestoJulienDuvignau]

Hi @ilvalerione !

Ok i get it for the VectorStore. Sorry i'm learning everything about LLM and AI Agents.
So the problem is, i can't get any answers from the document history.md. It's look like my agent didn't know anything about it.

[ilvalerione]

Since you are using Inspector don't you see the vector store search tasks in the agent timeline?

[TabestoJulienDuvignau]

I struggle to have consistent data in the inspector, most of the time i got no trace.
I figure to have one and i can see the vector search, but the answer from OpenAI LLM is not relevant at all.

[ilvalerione]

Drop in a live chat on the Inspector dashboard so we can take a look at the data.