https://github.com/theveryhim/chatbot

A chat agent equipped with RAG&Web search and capable of 20Q guessing game
https://github.com/theveryhim/chatbot

chatbot deep-learning gradio-interface llm natural-language-processing

Last synced: 4 months ago
JSON representation

A chat agent equipped with RAG&Web search and capable of 20Q guessing game

• Host: GitHub
• URL: https://github.com/theveryhim/chatbot
• Owner: theveryhim
• License: mit
• Created: 2025-09-14T13:53:25.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2025-09-14T22:55:15.000Z (9 months ago)
• Last Synced: 2025-09-15T00:26:43.912Z (9 months ago)
• Topics: chatbot, deep-learning, gradio-interface, llm, natural-language-processing
• Homepage:
• Size: 216 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# ChatBot [![Kaggle](https://kaggle.com/static/images/open-in-kaggle.svg)](https://www.kaggle.com/code/armanyazdani/chatbot)

Our goal here is to design and implement a conversational chatbot based on a Large 

Language Model (LLM) with the ability to preserve and utilize conversation history. Unlike a 

simple chatbot that only responds to the most recent input, our system maintains awareness of the 

entire dialogue context, enabling it to provide more coherent, context-aware, and human-like 

responses. This feature is crucial in creating natural and intelligent interactions where users may 

refer to earlier parts of the conversation.
 

*Find detailed explanations of any of the sections in notebook*



    

    
 7B-size chatbot 


## Context aware chat

- Initialization → Loads model and tokenizer. 

- User Input → Reads message from console. 

- History Update → Maintains all previous exchanges in conversation_history. 

- Context Management → Before generating, the history may be trimmed using manage_context_window. 

- Input Preparation → Combines history + current user input into tokenized format. 

- Response Generation → Calls the model to produce a reply. 

- Decoding → Converts raw tokens to human-readable response. 

- Display + Save → Prints the bot reply and updates history for the next turn.

## Retrieval-Augmented Generation (RAG) 



    

    
 Retrieved data from user's pdf 


- RAG System Initialization

    - initialize rag if possible

    - if PDF exists at PDF PATH then

      - Extract text using 

      - Split into chunks

      - Initialize embedding model

      - Generate embeddings and normalize

      - Create FAISS index 

      - Add embeddings to index

      - Set rag enabled 

-  Information Retrieval Process

    - if rag enabled AND index exists AND chunks not empty then

      - Encode query

      - Search index

      - Format results with reference numbering

      - Return formatted system message

## WebIntegration 



    

    
 Web search sample 


- Preserve original user input for display

- if web search enabled AND needed then

  - search results= searchexa(user input)

  - Update data_state.webdata

## Interactive Game



    

    
 20Q game view 


- Initialization 

  - Import ValidatorModel. 

  - Instantiate validator_model = ValidatorModel(). 

- Game Start 

  - Prompt the user (or system input) to confirm that the user intends to play. 

- Questioning Phase  

  - Generate a yes/no question about the secret word (e.g., "Is this an animal?"). 

  - Submit the question: validator_model.validate_question(question). 

  - Store the response. 

- Guessing Phase 

  - Use accumulated answers to form a hypothesis about the secret word. 

  - Submit the guess: validator_model.validate_guess(guess). 

  - If correct, terminate with success. 

- Termination 

  - If the model cannot guess correctly within 20 questions, terminate and mark as unsuccessful. 

## User Interface with Gradio

The interface implements a multi-tab architecture with sophisticated state management and real-time monitoring.



    

    
 Tabs and resources