Skip to content

rag-retrieval.md

Latest commit

 

History

History
128 lines (93 loc) · 3.65 KB

rag-retrieval.md

File metadata and controls

128 lines (93 loc) · 3.65 KB
Raw

RAG Retrieval & Pipeline Guide

Retrieval is the act of finding relevant context from your knowledge base to answer a user's question. This guide covers retrieval strategies and how to configure the RAG pipeline.

Retrieval Strategies

MobileGraph supports several built-in retrievers to balance speed and accuracy.

1. VectorStoreRetriever (Semantic Search)

The most common retriever. It uses vector embeddings to find documents with similar meanings, even if keywords don't match.

val retriever = VectorStoreRetriever(
    vectorStore = myStore,
    topK = 5,
    similarityThreshold = 0.75f // Only return highly relevant chunks
)

2. BM25Retriever (Keyword Search)

A traditional search engine algorithm. It is excellent for finding exact matches, technical terms, or specific product IDs that vector search might miss.

val retriever = BM25Retriever(allDocuments)

3. HybridRetriever (Balanced Search)

Combines Vector and BM25 search results using Reciprocal Rank Fusion (RRF). This is often the best choice for production apps.

val hybrid = HybridRetriever(listOf(vectorRetriever, bm25Retriever))

4. MultiQueryRetriever (Improved Recall)

Uses an LLM to generate variations of the user's question before searching. This helps when the initial query is vague.

val multiQuery = MultiQueryRetriever(vectorRetriever, chatModel)

Targeted Retrieval (Filtering)

You can restrict your search to a specific document or subset of data using RetrievalFilter.

// Only search within the specific user manual
val filter = RetrievalFilter.DocumentId("manual-001")

val results = retriever.retrieve("how to reset?", filter = filter)

The RAG Pipeline

The RagPipeline orchestrates the entire process: Search -> Augment -> Generate.

Basic Setup

MobileGraph.initialize() {
    // ... model and vector store config ...

    withRetrievers {
        register("default", vectorRetriever, isDefault = true)
    }

    withRag {
        retriever(vectorRetriever)
        chatModel(myChatModel)
        // contextBuilder(customBuilder) // Optional: for custom prompts/guardrails
    }
}

Executing a Query

// 1. Standard execution (returns model response)
val result = MobileGraph.rag.pipeline().execute("What is the battery life?")

// 2. Detailed execution (includes debug metadata)
val detailed = MobileGraph.rag.pipeline().executeDetailed("...")
println("Prompt used: ${detailed.finalPrompt}")
println("Scores: ${detailed.retrievedDocs.map { it.score }}")

Creating a Custom Retriever

Implement the Retriever interface to integrate any external search engine or custom logic.

class MyCustomRetriever : Retriever {
    override suspend fun retrieve(
        query: String,
        filter: RetrievalFilter?,
        context: ExecutionContext
    ): List<Document> {
        // Your custom logic: call a REST API, query a DB, etc.
        return listOf(Document(id = "ext-1", content = "External content..."))
    }
}

Customizing Prompts & Guardrails

You can control how context is presented to the LLM and add strict instructions by providing a custom ContextBasedPromptBuilder.

class GuardrailedBuilder : ContextBasedPromptBuilder {
    override fun buildContextAndPrompt(documents: List<Document>?, userQuery: String): String {
        val contextText = documents?.joinToString("\n") { it.content } ?: ""
        return """
            Answer using ONLY the context below. 
            If you don't know, say "Not enough info".
            
            Context: $contextText
            Question: $userQuery
        """.trimIndent()
    }
}