ChatGSPP

Overview

ChatGSPP is a privacy-focused AI chatbot designed to answer questions based on the content of a specific website. Built on top of PrivateGPT, it leverages a Retrieval-Augmented Generation (RAG) pipeline to ensure that responses are contextually relevant to the ingested website content.

Unlike traditional chatbots that rely on external knowledge sources, ChatGSPP is fully self-contained, meaning it primarily references content from the specified website to generate responses. This ensures both accuracy and privacy while enabling organizations to provide a website-specific AI assistant.

---

How It Works

ChatGSPP enhances PrivateGPT by implementing a website-focused workflow with the following steps:

1️⃣ Document Ingestion for Website Content

• Parses and ingests data from the target website (e.g., pages, articles, FAQs, etc).
• Supports web scraping or direct document uploads to convert website content into structured text.
• Uses embedding generation to transform each content section into vector embeddings for retrieval.

2️⃣ Contextual Search and Retrieval

• When a user submits a query:
  • The query is converted into an embedding.
  • The system searches for relevant website sections in a vector database (e.g., Qdrant).
  • The most relevant snippets are retrieved to ensure responses are based on the website’s actual content.

3️⃣ Enhanced Prompt Engineering

• The retrieved information is formatted into a structured prompt.
• This prompt is fed into the LLM (e.g., Llama3.1, DeepSeek R1, Gemma 2) to generate a response.
• Responses remain accurate, domain-specific, and context-aware.

4️⃣ PrivateGPT-Powered Response Generation

• ChatGSPP leverages PrivateGPT’s API to generate responses while ensuring privacy.
• The LLM processes the contextual prompt and generates a coherent, website-specific answer.
• The response is displayed in the chatbot interface.

5️⃣ Web Chatbot Interface

• ChatGSPP features a user-friendly web-based chat interface.
• The chatbot interacts with the PrivateGPT-powered backend to handle queries dynamically.

---

Why Use ChatGSPP?

• Website-Specific Answers → Responses are strictly based on website content, reducing misinformation.
• Dynamic Updates → As the website evolves, new content is ingested and indexed.
• Privacy-Focused → Runs locally or in a controlled environment, ensuring sensitive website data is never shared externally.
• Customizable → Swap out LLMs (e.g., LlamaCPP, OpenAI, GPT4All) and vector databases as needed.
• Scalable & Efficient → Uses dependency injection and modular abstractions to make component swapping and upgrades seamless.
• Open-Source Support → Under the Apache 2.0 license, continuing the tradition of open innovation and collaboration.

---

🏗 Architecture Overview

ChatGSPP builds upon PrivateGPT’s RAG pipeline while tailoring it to website-based question answering.

Core Components:

1. API Layer (FastAPI)

   • Manages user requests via a router-service architecture.
   • Interfaces with the RAG pipeline and document embeddings.

2. RAG Pipeline

   • Uses LlamaIndex abstractions for document ingestion, search, and prompt engineering.
   • Retrieves relevant website content based on user queries.

3. Component Abstractions

   • LLM → Supports models like LlamaCPP or OpenAI.
   • BaseEmbedding → Generates vector embeddings from website content.
   • VectorStore → Stores and retrieves document embeddings (e.g., Qdrant, FAISS).

4. Dependency Injection

   • Allows swapping out LLMs, embedding models, and vector stores without modifying core logic.

5. Web Chat Interface

   • Provides a frontend for users to ask questions and receive real-time, website-specific answers.

---

🚀 Getting Started

Reference the official PrivateGPT docs to get started with a fully local PrivateGPT setup. From there, you can reference chatGSPP's build docs to replicate chatGSPP's configuration on your home computer.

---

📜 License

ChatGSPP is open-source and licensed under the Apache 2.0 license. Feel free to use, modify, and contribute!

---

🙌 Contributions & Support

Got ideas? Found a bug? Feel free to submit an issue or a pull request!

• GitHub Issues: Report Bugs
• Discussions: Join the Community

Partners & Supporters

PrivateGPT is actively supported by the teams behind:

• Qdrant: Providing the default vector database.
• Fern: Offering Documentation and SDKs.
• LlamaIndex: Supplying the base RAG framework and abstractions.

This project is strongly influenced and supported by other amazing projects such as:

• LangChain
• GPT4All
• LlamaCpp
• Chroma
• SentenceTransformers

Contact

For any questions, suggestions, or contributions, please reach out to GSPP's IT Team:

• Email: [email protected]
• GitHub Issues: ChatGSPP Issues