README.md

---
title: Document AI Agent
emoji: 💬
colorFrom: yellow
colorTo: purple
sdk: gradio
sdk_version: 4.36.1
app_file: app.py
pinned: false
license: apache-2.0
---

# NVIDIA AI Document Chatbot

This project is a document-based chatbot application. It helps users ask questions about specific documents and receive accurate responses based on those documents.

## Models and Components Used:

1. **NVIDIAEmbeddings (NV-Embed-QA)**:
   - This model extracts vector representations of texts to better understand documents.
   - The **NV-Embed-QA** model is used to find relevant information in documents to answer questions.

2. **ChatDocument (mistralai/mixtral-8x7b-instruct-v0.1)**:
   - The **Mistral-8x7B Instruct** model is responsible for answering user questions about documents. It specializes in extracting information from documents and responding conversationally.

## Application Workflow:

1. **Loading Documents**:
   - Specific academic papers are loaded using `ArxivLoader`. These documents are split into text chunks and filtered based on predefined rules.
   - The documents are then added to a FAISS vector store, which allows for efficient and fast document chunk retrieval.

2. **Chat and Document Querying**:
   - The user's questions are processed according to a predefined chat template. The response is generated based on both the conversation history and information retrieved from the documents.
   - The `chat_gen` function takes the user's input and generates responses using NVIDIA models, pulling relevant information from the documents.

3. **Remembering Document Content and Conversation History**:
   - Previous user messages and responses are stored in a conversation memory and used for answering future questions more effectively.

### Conclusion:

This application leverages NVIDIA's powerful language models and embedding tools to generate intelligent, document-driven conversational responses.


An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).