Create a Knowledge-Based Q&A Application with DeepSeek and Tencent Cloud

Large language models (LLMs) are revolutionizing many fields, including knowledge management. With the help of Tencent Cloud's Large Model Knowledge Engine (LKE), you can now create knowledge-based Q&A applications powered by the DeepSeek model, enabling you to leverage your private knowledge documents for providing intelligent answers.

This article walks you through the process of building a DeepSeek-R1/V3-based knowledge Q&A application utilizing the full potential of the 671B DeepSeek model and Retrieval Augmented Generation (RAG) on Tencent Cloud.

Understanding the Power of DeepSeek and RAG

The combination of DeepSeek and RAG offers a powerful solution for developing intelligent knowledge-based applications.

• DeepSeek-R1/V3: A large language model known for its strong understanding and generation capabilities.
• RAG (Retrieval Augmented Generation): A framework for enhancing LLMs with external knowledge sources. This approach improves accuracy and relevance by grounding the model's responses in retrieved documents.

Step-by-Step Guide to Creating a DeepSeek Knowledge Q&A Application

Here’s a detailed guide to setting up your own DeepSeek knowledge Q&A application using Tencent Cloud:

1. Account Setup and Product Activation:

• Log in: Access your Tencent Cloud account or register for a new one if you don't have one.
• Activate the Knowledge Engine: Navigate to the Large Model Knowledge Engine product page and activate the service.

2. Application Creation:

• Access the Application Management: In the Experience Center, find and click "Application Management" in the left menu to create a new application.
• Basic Information: Set the application's name, avatar and other basic details. After saving, this information will undergo a security review.

3. Knowledge Management:

• Access Knowledge Management: Once your application is created, enter it and click "Knowledge Management" to begin importing your private knowledge documents.
• Document Upload: Upload relevant documents such as product manuals. The platform supports various formats, including:
  • PDF, DOC, DOCX, PPT, PPTX (up to 200MB)
  • XLSX, XLS, MD, TXT, CSV (up to 20MB)
  • PNG, JPG, JPEG images with text (up to 50MB, aspect ratio limit 1:7)
• Document Processing: The platform will parse the documents and store them in a knowledge base.
• Document Management: The platform allows managing document-based knowledge, where one can check document character count, update time, state. It also supports setting document expiration date, downloading documents, and deleting documents.

4. Application Configuration:

• Icon and Name: Click Application Avatar in the application configuration interface where you can change the application's icon and name.
• Model Selection: Choose DeepSeek-R1 from the options available in Generate Model List, and enable both the "Document" and "Q&A" switches.
• Answer Settings: Enhance response accuracy by disabling "Internet Search" and setting the reply behavior for questions outside the knowledge base to "Reply according to filled-in content".

5. Testing and Validation:

• Dialogue Testing: Enter your questions in the dialogue test window to check that the application will produce relevant and exact responses.
• Reference Source Review: Click specified resources from the documents used to respond to a query by clicking "Reference Source".

6. Deployment and Access:

• Publish Application: Select "Publish and Confirm", which graduates the developed environment to the Production environment.
• Access Information: Obtain the experience link, sharing link, and QR code via "Call Info" found when you click "Publish Management", after publication.
• API Integration: Use the API key, also found in Publish Management > Call Info, to integrate the service into your business systems via interface calls.

Optimizing Your Knowledge Base

• Document Format: Adhere to the format and size limitations for uploaded documents to ensure successful processing.
• Table Formatting: For spreadsheet files, organize data into single tables per sheet, and avoid empty rows to maintain answer quality.
• Knowledge Base Scope: Define the knowledge base scope to refine search results.

Conclusion

With Tencent Cloud's Large Model Knowledge Engine and the DeepSeek model, building a knowledge-based Q&A application is now easier than ever. By following these steps, you can create a powerful tool that leverages your private knowledge assets to provide intelligent and accurate answers to your users. Leverage the power of LLMs to enhance productivity, improve customer service, and unlock the full potential of your organization's collective knowledge.