Creating a chatbot with contextual retrieval using Cohere comma… by AMAH DanielCreating a chatbot with contextual retrieval using Cohere comma… by AMAH Daniel

Creating a chatbot with contextual retrieval using Cohere comma…

AMAH Daniel

Completed work

AI Application Developer

Cohere

Python

Streamlit

Artificial Intelligence

Project Overview

Chatish is an innovative Streamlit web application that demonstrates the power of contextual retrieval using large language models, specifically Cohere's Command R model. This project showcases how modern AI can transform document interaction through intelligent, context-aware conversations.

Architectural Components

The application is structured around four primary Python modules:

app.py: Main application entry point

chat_manager.py: Manages chat interactions

cohere_client.py: Handles AI interactions

file_handler.py: Processes uploaded documents

Application Architecture Diagram

graph TD
    A[User Interface - Streamlit] --> B[File Upload]
    A --> C[Chat Input]
    B --> D[File Handler]
    C --> E[Chat Manager]
    D --> F[Cohere Client]
    E --> F
    F --> G[AI Response Generation]
    G --> A

Key Implementation Details

File Handling Strategy

The FileHandler class demonstrates a flexible approach to document processing:

def process_file(self, uploaded_file):
    if uploaded_file.type == "application/pdf":
        return self.extract_text_from_pdf(uploaded_file)
    else:
        # Extensible for future file type support
        return uploaded_file.read().decode()

Intelligent Prompt Engineering

The CohereClient builds context-aware prompts:

def build_prompt(self, user_input, context=None):
    context_str = f"{context}\n\n" if context else ""
    return (
        f"{context_str}"
        f"Question: {user_input}\n"
        f"Provide straight to the point response except when told "
        f"to elaborate using available metrics and historical data"
    )

Conversation Management

The chat management includes intelligent history tracking:

def chat(self, user_input, context=None):
    # Maintain conversation history
    self.conversation_history.append({"role": "user", "content": user_input})

    # Limit history to prevent context overflow
    if len(self.conversation_history) > 10:
        self.conversation_history = self.conversation_history[-10:]

Technical Challenges Addressed

Context Retrieval: Dynamically incorporating uploaded document context

Conversation Persistence: Maintaining conversational state

Streaming Response: Implementing real-time AI response generation

Technology Stack

Web Framework: Streamlit

AI Integration: Cohere Command R

Document Processing: PyPDF2

Language: Python 3.9+

Performance Considerations

Token Limits: Configurable with max_tokens parameter

Temperature Control: Adjustable response creativity via temperature

Model Flexibility: Easy model swapping in configuration

Future Roadmap

Enhanced error handling

Support for additional file types

Advanced context chunking

Sentiment analysis integration

Deployment Considerations

Requirements

cohere==5.13.11
streamlit==1.41.1
PyPDF2==3.0.1

Quick Start

# Create virtual environment
python3 -m venv chatish_env

# Activate environment
source chatish_env/bin/activate

# Install dependencies
pip install -r requirements.txt

# Run application
streamlit run app.py