Build Department-Specific Internal Chat Interfaces with Needle and Haystack

In today's fast-paced work environment, having quick and easy access to information is crucial for productivity. With Needle's integration with Haystack, you can build tailored internal chat interfaces that cater to the unique needs of your department. Whether it's for HR, legal, or sales, this integration streamlines how your team interacts with vital documents and data.

What You Need to Get Started

To harness the power of Needle within your Haystack projects, you'll need to set up the Needle Document Store and Needle Embedding Retriever components. These tools make it simple to create a robust Retrieval-Augmented Generation (RAG) pipeline that can process queries and return contextually relevant answers from your organization's documents.

Installation

Begin by installing the Needle-Haystack package via pip:

pip install needle-haystack-ai

API Keys

Before diving into the integration, ensure you have your API keys ready. You can get them from the developer settings page. You will need to set the following environment variables:

• NEEDLE_API_KEY: Obtain this key from your Developer settings.

• OPENAI_API_KEY: This key is necessary for connecting to the OpenAI generator.

Example Pipeline: Building Your Chat Interface

To create an internal chat interface, we'll build a simple RAG pipeline using Needle's tools alongside an OpenAI generator. Below is a step-by-step guide:

Step 1: Set Up Your Document Store

In Needle, document stores are referred to as collections. Start by creating a reference to your Needle collection using the NeedleDocumentStore class.

from needle_haystack import NeedleDocumentStore, NeedleEmbeddingRetriever

document_store = NeedleDocumentStore(collection_id="<ID>")

retriever = NeedleEmbeddingRetriever(document_store=document_store)

Step 2: Define the Prompt for the LLM

Next, we will craft a prompt template that will guide the OpenAI generator in formulating responses based on the retrieved documents.

from haystack import Pipeline
from haystack.components.generators import OpenAIGenerator
from haystack.components.builders import PromptBuilder

prompt_template = """
Given the following retrieved documents, generate a concise and informative answer to the query:

Query: {{query}}
Documents:
{% for doc in documents %}
    {{ doc.content }}
{% endfor %}

Answer:
"""

prompt_builder = PromptBuilder(template=prompt_template)
llm = OpenAIGenerator()

Step 3: Assemble Your RAG Pipeline

Now that you have your components ready, let's assemble the pipeline.

# Create a new Haystack pipeline
pipeline = Pipeline()

# Add components to the pipeline
pipeline.add_component("retriever", retriever)
pipeline.add_component("prompt_builder", prompt_builder)
pipeline.add_component("llm", llm)

# Connect the components
pipeline.connect("retriever", "prompt_builder.documents")
pipeline.connect("prompt_builder", "llm")

Step 4: Run the Pipeline

With everything set up, you can now run your RAG pipeline to answer user queries.

# Example query
prompt = "What are the latest updates on the company policies?"

result = pipeline.run({
    "retriever": {"text": prompt},
    "prompt_builder": {"query": prompt}
})

# Print the final answer
print(result['llm']['replies'][0])

Step 5: Tailor for Your Department

The beauty of this integration is that you can customize it according to your department's needs. Whether it's specific HR queries, project documentation in development, or customer support inquiries, the RAG pipeline can be adapted to fetch and present the most relevant information, enhancing your team's efficiency.

Support and Further Guidance

For more detailed guides and comprehensive support, please visit our documentation. If you have questions or feature requests, don't hesitate to connect with us on our Discord channel.