Replace dummy LLM-generated RAG with real vector database integration

[Itz-Agasta]

The current RAG agent implementation in apps/api-gateway/src/agents/rag-agent.ts uses a dummy approach where it prompts an LLM (Groq) to generate mock research paper data instead of performing actual semantic search against a vector database.

Current Implementation Analysis

// Current dummy implementation
const { text: summary } = await generateText({
  model: groq(config.models.ragAgent),
  system: `You are a research librarian... Generate a list of relevant research papers...`,
});

Issues:

• Uses LLM as a mock data generator
• No vector embeddings
• No persistent storage
• No real semantic search
• Inconsistent and potentially inaccurate results