RAG Engine

// RAG Pipeline Implementation
class RAGSystem(private val openAI: OpenAI, private val supabase: Supabase) {
    
    suspend fun query(userQuery: String): RAGResponse {
        // Generate query embedding
        val queryEmbedding = openAI.createEmbedding(userQuery)
            .data.first().embedding
        
        // Retrieve relevant documents
        val relevantDocs = supabase.matchDocuments(
            embedding = queryEmbedding,
            threshold = 0.78,
            count = 3
        )
        
        // Build context from retrieved documents
        val context = relevantDocs.joinToString("\n\n") {
            "[${it.metadata["source"]}]: ${it.content}"
        }
        
        // Generate contextual response
        val response = openAI.chatCompletion(listOf(
            Message("system", "Use only this context: $context"),
            Message("user", userQuery)
        ))
        
        return RAGResponse(
            answer = response.choices.first().message.content,
            sources = relevantDocs.map { it.metadata["source"] as String }
        )
    }
}