Constrained Generation in Retrieval-Augmented Generation Systems

Introduction

Welcome to the first lesson of the " Beyond Basic RAG: Improving our Pipeline " course, part of the " Foundations of RAG Systems " course path! In previous courses, you delved into the basics of Retrieval-Augmented Generation (RAG), exploring text representation with a focus on embeddings and vector databases. In this course, we'll embark on an exciting journey to enhance our RAG systems with advanced techniques. Our focus in this initial lesson is on constrained generation, a powerful method to ensure that language model responses remain anchored in the retrieved context, avoiding speculation or unrelated content. Get ready to elevate your RAG skills and build more reliable systems!

Theoretical Foundations of Constrained Generation

When employing large language models (LLMs) in real-world applications, accuracy and fidelity to a trusted dataset are paramount. Even advanced LLMs can produce incorrect or fabricated information — often termed “hallucinations.” This is where constrained generation becomes indispensable. In essence, it is a form of advanced prompt engineering : we carefully craft instructions so the LLM only responds using the retrieved information or provides disclaimers when insufficient data is found. By shaping the prompt and enforcing rule-based fallback mechanisms, we instruct the LLM to: Use only the data you supply (the “retrieved context”). Provide disclaimers or refusal messages when context is insufficient. Optionally cite which part of the content it used. The result is a system less prone to made-up facts and more consistent with the original knowledge source.

Why Constrained Generation Is Important

LLM hallucination can be quite misleading. Imagine a scenario where your application confidently presents policies or regulations not present in your knowledge base. This can create confusion or even compliance issues. With constrained generation : The model remains grounded in the retrieved context only. Uncertain or unavailable information triggers a fallback message like “No sufficient data.” You can require the model to cite lines to verify the source of the answer, building trust with users.

Defining the Constrained Generation Function

Join the 1M+ learners on CodeSignal

Be a part of our community of 1M+ users who develop and demonstrate their skills on CodeSignal

public static String[] generateWithConstraints(String query, String retrievedContext, String strategy) { /** * Thoroughly enforce model reliance on 'retrievedContext' when answering 'query'. * * The 'strategy' parameter allows for different prompt template variations: * 1) Base approach: Provide context, instruct LLM not to use outside info, * and respond with 'No sufficient data' if the context is insufficient. * 2) Strict approach: Provide context with explicit disclaimers if the answer is not found. * 3) Citation approach: Provide context, then request the LLM to cite the relevant lines. * * Robust fallback: * - If 'retrievedContext' is empty, respond with an apology or neutral statement. * - Optionally log each stage for debugging or performance analysis. */ // Provide a safe fallback if no context is retrieved if (retrievedContext.trim().isEmpty()) { return new String[]{"I'm sorry, but I couldn't find any relevant information.", "No context used."}; } String prompt; // Choose a prompt template based on strategy if (strategy.equals("base")) { // Base approach prompt = ( "Use the following context to answer the question in a concise manner.\n\n" + "Context:\n" + retrievedContext + "\n" + "Question: '" + query + "'\n" + "Answer:" ); } else if (strategy.equals("strict")) { // Strict approach: explicitly disallow info beyond the provided context prompt = ( "You must ONLY use the context provided below. If you cannot find the answer in the context, say: 'No sufficient data'.\n" + "Do not provide any information not found in the context.\n\n" + "Context:\n" + retrievedContext + "\n" + "Question: '" + query + "'\n" + "Answer:" ); } else if (strategy.equals("cite")) { // Citation approach: require references to lines used prompt = ( "Answer strictly from the provided context, and list the lines you used as evidence with 'Cited lines:'.\n" + "If the context does not contain the information, respond with: 'Not available in the retrieved texts.'\n\n" + "Provided context (label lines as needed):\n" + retrievedContext + "\n" + "Question: '" + query + "'\n" + "Answer:" ); } else { prompt = "Invalid strategy specified."; } // Print the prompt for debugging or inspection System.out.println("Prompt: \n " + prompt + "\n"); // Make call to the LLM String response = LLM.getLlmResponse(prompt); // Attempt to parse out 'Cited lines:' if present String[] segments = response.split("Cited lines:"); if (segments.length == 2) { String answerPart = segments[0].trim(); String usedContextPart = segments[1].trim(); return new String[]{answerPart, usedContextPart}; } else { // If the LLM didn't provide citations, treat the entire response as the answer return new String[]{response.trim(), "No explicit lines cited."}; } }

Demonstration of Retrieval and Constrained Generation

Below is a typical scenario. We assume documents have been chunked and stored in a vector database. We only show the retrieval steps briefly:

Java

public static void main(String[] args) throws IOException, ApiException {
    // Example usage demonstrating retrieval followed by constrained generation

    // 1. Get preloaded collection
    Collection collection = VectorDB.getCollection("full_collection");

    // 2. Run a sample query
    String query = "Highlight the main effects of AI";
    QueryResponse retrievalResults = collection.query(Arrays.asList(query), 2, null, null, null);

    // 3. Construct the retrieved context from top matches
    String retrievedContext = "";
    if (retrievalResults.getDocuments().size() > 0 && retrievalResults.getDocuments().get(0).size() > 0) {
        StringBuilder contextBuilder = new StringBuilder();
        for (String docText : retrievalResults.getDocuments().get(0)) {
            contextBuilder.append("- ").append(docText).append("\n");
        }
        retrievedContext = contextBuilder.toString().trim();
    }

    // 4. Execute constrained generation function for demonstration
    for (String strategy : new String[]{"base", "strict", "cite"}) {
        String[] result = generateWithConstraints(query, retrievedContext, strategy);
        String answer = result[0];
        String usedContext = result[1];
        
        System.out.println("Strategy: " + strategy);
        System.out.println("Constrained generation answer:\n" + answer);
        System.out.println("Context or lines used: " + usedContext + "\n");
    }
}

public static void main(String[] args) throws IOException, ApiException {
    // Example usage demonstrating retrieval followed by constrained generation

    // 1. Get preloaded collection
    Collection collection = VectorDB.getCollection("full_collection");

    // 2. Run a sample query
    String query = "Highlight the main effects of AI";
    QueryResponse retrievalResults = collection.query(Arrays.asList(query), 2, null, null, null);

    // 3. Construct the retrieved context from top matches
    String retrievedContext = "";
    if (retrievalResults.getDocuments().size() > 0 && retrievalResults.getDocuments().get(0).size() > 0) {
        StringBuilder contextBuilder = new StringBuilder();
        for (String docText : retrievalResults.getDocuments().get(0)) {
            contextBuilder.append("- ").append(docText).append("\n");
        }
        retrievedContext = contextBuilder.toString().trim();
    }

    // 4. Execute constrained generation function for demonstration
    for (String strategy : new String[]{"base", "strict", "cite"}) {
        String[] result = generateWithConstraints(query, retrievedContext, strategy);
        String answer = result[0];
        String usedContext = result[1];
        
        System.out.println("Strategy: " + strategy);
        System.out.println("Constrained generation answer:\n" + answer);
        System.out.println("Context or lines used: " + usedContext + "\n");
    }
}

Under the hood:

We load the corpus, build a vector collection, and issue a query.
The top two documents are retrieved, combined, and passed to the constrained generation function.

Introduction

Theoretical Foundations of Constrained Generation

Why Constrained Generation Is Important

Defining the Constrained Generation Function

Join the 1M+ learners on CodeSignal

Generating the Final Response

Demonstration of Retrieval and Constrained Generation

Practical Example: A Policy FAQ Bot

Conclusion and Next Steps