Skip to main content

Retrieval QA Chain

Overview​

The Retrieval QA Chain in AnswerAI is a specialized chain designed to answer questions based on retrieved documents. Unlike the Conversational Retrival QA Chain, this chain focuses on providing accurate answers from a knowledge base without maintaining conversation history.

Key Differences from Conversational Chain​

  1. No Memory: The Retrieval QA Chain does not maintain memory of previous interactions. Each query is treated independently.

  2. Document Retrieval: This chain actively retrieves relevant documents for each query, whereas the Conversational Retrival QA Chain relies on its conversation history.

  3. Single-Turn Interactions: Optimized for standalone questions rather than multi-turn conversations.

  4. Knowledge-Base Focused: Answers are derived from a specific knowledge base rather than general knowledge or conversation context.

Key Benefits​

  • Accuracy: Provides answers based on specific, retrieved information.
  • Scalability: Can handle a large knowledge base efficiently.
  • Consistency: Gives consistent answers to similar questions, regardless of conversation history.
  • Flexibility: Can be easily updated with new information by updating the underlying vector store.

When to Use Retrieval QA Chain​

Use this chain when:

  1. You need to answer questions based on a specific knowledge base or dataset.
  2. Conversation history is not important or relevant.
  3. You want to provide factual, consistent answers across different user sessions.
  4. Dealing with domain-specific queries that require accurate, up-to-date information.

How It Works​

  1. Query Reception: The chain receives a question via the AnswerAI API.
  2. Document Retrieval: Relevant documents are fetched from the vector store.
  3. Context Preparation: Retrieved documents are prepared as context for the language model.
  4. Answer Generation: The language model generates an answer based on the question and retrieved context.
  5. Response Delivery: The answer is returned through the API.

Key Components​

  1. Language Model: Processes the query and generates answers.
  2. Vector Store Retriever: Fetches relevant documents based on the query.
  3. Input Moderation (Optional): Ensures safe and appropriate inputs.

Using the Retrieval QA Chain with AnswerAI API​

Basic Usage​

import requests

API_URL = "http://localhost:3000/api/v1/prediction/<your-chatflowid>"
API_KEY = "your-api-key"

headers = {
    "Authorization": f"Bearer {API_KEY}",
    "Content-Type": "application/json"
}

data = {
    "question": "What are the main features of quantum computing?",
    "overrideConfig": {
        "vectorStoreRetriever": {
            "type": "pinecone",
            "indexName": "quantum-computing-index"
        }
    }
}

response = requests.post(API_URL, json=data, headers=headers)
print(response.json())

Adding Input Moderation​

data = {
    "question": "What are the main features of quantum computing?",
    "overrideConfig": {
        "vectorStoreRetriever": {
            "type": "pinecone",
            "indexName": "quantum-computing-index"
        },
        "inputModeration": [
            {
                "type": "toxicity",
                "threshold": 0.8
            }
        ]
    }
}

response = requests.post(API_URL, json=data, headers=headers)
print(response.json())

Use Case Example: Technical Support System​

import requests

API_URL = "http://localhost:3000/api/v1/prediction/<your-chatflowid>"
API_KEY = "your-api-key"

def technical_support(query):
    data = {
        "question": query,
        "overrideConfig": {
            "vectorStoreRetriever": {
                "type": "pinecone",
                "indexName": "technical-support-docs"
            }
        }
    }

    response = requests.post(API_URL, json=data, headers={
        "Authorization": f"Bearer {API_KEY}",
        "Content-Type": "application/json"
    })
    return response.json()['text']

# Usage
print(technical_support("How do I reset my router?"))
print(technical_support("What are the system requirements for the latest software update?"))

Best Practices​

  1. Quality of Vector Store: Ensure your vector store contains high-quality, relevant documents.
  2. Regular Updates: Keep your knowledge base up-to-date for accurate answers.
  3. Clear Queries: Encourage users to ask clear, specific questions for better retrieval.
  4. Fallback Mechanisms: Implement fallbacks for when the chain can't find relevant information.
  5. Monitoring: Regularly review the chain's performance and the relevance of retrieved documents.

Limitations and Considerations​

  • No Conversation Context: Unlike the Conversational Chain, this chain doesn't consider previous interactions.
  • Retrieval Dependence: The quality of answers heavily depends on the quality and relevance of retrieved documents.
  • Single-Query Focus: May not perform as well for complex queries that require multi-step reasoning.

By leveraging the Retrieval QA Chain in AnswerAI, you can create powerful question-answering systems that provide accurate, knowledge-based responses without the need for maintaining conversation history. This makes it ideal for scenarios where factual, consistent information retrieval is more important than contextual conversation.