Leveraging AI for Smarter Information Retrieval: Circles' Expertise in Building RAG Pipelines

Retrieval-Augmented Generation (RAG) is an AI approach that enhances the accuracy and contextual understanding of large language models (LLMs) by integrating external knowledge retrieval. Instead of relying solely on a language model’s pre-trained knowledge, a RAG pipeline retrieves relevant content from a database or document repository and feeds it as context to the model before generating a response. This significantly improves the relevance, accuracy, and depth of responses.

Organizations deal with multiple formats of information, including PDFs, Word documents, and video content. Extracting, structuring, and retrieving information from these sources in real time is a complex challenge. Circles’ expertise lies in building scalable RAG pipelines that process both text and video-based knowledge efficiently, ensuring seamless AI-driven search experiences.

1. Extracting Text from Documents and Videos

Circles’ AI solutions use state-of-the-art Natural Language Processing (NLP) models to extract and structure textual data from various sources:

• PDFs & Documents: Libraries like pdfplumber and python-docx extract text from PDFs and Word files.
• Video Transcriptions: Speech-to-text models like OpenAI’s Whisper transcribe video and audio content into structured text.

2. Generating Embeddings for Efficient Retrieval

Circles leverages cutting-edge embedding models like sentence-transformers to convert extracted text into high-dimensional vector representations. These embeddings help in identifying semantically relevant content efficiently.

from sentence_transformers import SentenceTransformer

model = SentenceTransformer('all-MiniLM-L6-v2')
def generate_embeddings(text_chunks):
    return model.encode(text_chunks, convert_to_tensor=True)

3. Storing and Searching Embeddings with FAISS

To enable rapid retrieval, Circles integrates FAISS, a scalable similarity search engine, to store and search vector embeddings.

import faiss, numpy as np

embedding_dim = 384
index = faiss.IndexFlatL2(embedding_dim)
metadata = []

def add_to_index(text_chunks):
    embeddings = generate_embeddings(text_chunks)
    index.add(np.array(embeddings))
    metadata.extend(text_chunks)

def search_index(query, top_k=5):
    query_embedding = generate_embeddings([query])[0].unsqueeze(0).cpu().numpy()
    distances, indices = index.search(query_embedding, top_k)
    return [metadata[i] for i in indices[0]]

4. Integrating with a Local LLM for Response Generation

To maintain data privacy and reduce reliance on cloud APIs, Circles integrates RAG pipelines with local AI models like LLaMA, Mistral, or Falcon using the transformers library.

from transformers import pipeline

generator = pipeline("text-generation", model="path/to/local-model")

def generate_response(prompt):
    response = generator(prompt, max_length=150, num_return_sequences=1)
    return response[0]['generated_text']

5. Deploying as an API for Seamless Access

Circles ensures that AI-driven search capabilities are accessible through intuitive APIs, allowing organizations to integrate RAG pipelines into their workflows.

from fastapi import FastAPI

app = FastAPI()

@app.get("/query/")
def query_pipeline(query: str):
    relevant_chunks = search_index(query)
    context = " ".join(relevant_chunks)
    prompt = f"Context: {context}\n\nQuestion: {query}\nAnswer:"
    response = generate_response(prompt)
    return {"response": response}

Circles’ expertise in AI and NLP enables businesses to:

• Enhance search accuracy by leveraging real-time knowledge retrieval.
• Optimize content accessibility across multiple formats, including text and video.
• Maintain data privacy by deploying on-premise AI solutions.
• Scale RAG pipelines to support enterprise-level workloads seamlessly.

Whether your organization needs an AI-powered search tool for internal knowledge management, customer support automation, or content discovery, Circles has the expertise to deliver customized, efficient, and high-performing RAG solutions.

Please contact sales for further information:

Teo Kohonen
teo.kohonen@circles.fi

+358 10 3843 330