Company-Wide-AI-Assistant (Version-2.1.0)

Description

A sophisticated AI assistant that provides department-specific information by automatically scanning and processing company documents from Google Drive. Uses Ollama's local language models with direct integration to your organization's document management system.

✨ What's New in version 2.1

🧠 RNN-Powered Department Classification

Intelligent department detection using Recurrent Neural Networks:

• Mathematical RNN equations: h_t = tanh(Whx*x_t + Whh*h_{t-1} + b_h)
• Enhanced classification: Combines RNN speed with LLM semantic understanding
• Confidence scoring: Automatic threshold-based department selection
• Continuous learning: Improves from user interactions over time

🔄 Smart Multi-Department Querying

Ask questions once, get answers from all relevant departments:

• Automatic department detection: No need to specify department manually
• Confidence-based filtering: Only queries departments above threshold
• Combined insights: Presents answers from multiple relevant departments
• Visual confidence indicators: Color-coded confidence levels (🟢🟡🔴)

💾 Persistent Model Training

Model improvements are saved and reused:

• Auto-save functionality: Best models automatically saved during training
• Training history: Track performance improvements over time
• Vocabulary persistence: Custom vocabulary built from training data
• Model status tracking: View training date, accuracy, and sample count

📚 Continuous Learning System

The system learns and improves from usage:

• Interaction logging: All queries and answers stored for training
• Automatic retraining: Triggers after every 5 interactions
• Incremental learning: Adds new unique questions to training set
• Performance monitoring: Accuracy tracking and model validation

🎯 Existing Features from Previous Versions

🔗 Google Drive Integration

Seamlessly connect your workspace with Google Drive.

• Automatic authentication with OAuth 2.0
• Secure token persistence using encrypted pickle files
• Direct access to company documents without manual downloads

---

📁 Multi-Format Document Support

Enhanced compatibility for diverse document types:

• Google Docs
• Microsoft Word (.docx)
• PDF Documents
• Plain Text Files

---

🔍 Smart Document Discovery

Smarter automation for report and directory management:

• Automatic folder scanning for "Company Reports" directory
• Department-specific detection (Finance, Marketing, IT)
• Weekly report recognition via Week-XX pattern
• File type filtering and validation for clean processing

---

💾 In-Memory Processing

Optimized for performance and security:

• No local file storage — all operations handled in memory
• Memory-efficient processing for large documents
• Secure handling without temporary files

---

🚀 Version 2.1 marks a major step toward custom RNN, smarter multiple department query, and enhanced fall backs.

Requirements

• Python 3.6+
• PyTorch (for RNN functionality)
• Ollama installed locally
• Llama 3.1:8B model (or compatible model)
• Google Cloud Project with Drive API enabled
• Google OAuth 2.0 credentials

Installation

1. Install Ollama

# Visit https://ollama.ai/download and follow installation instructions for your OS
# Or use one of these commands:

# macOS
# curl -fsSL https://ollama.ai/install.sh | sh

# Windows
# Download from https://ollama.ai/download

# Linux
# curl -fsSL https://ollama.ai/install.sh | sh

2. Pull the Required Model

ollama pull llama3.1:8b

3. Install Python Dependencies

pip install ollama-python
pip install ollama-python google-auth-oauthlib google-auth-httplib2 google-api-python-client
pip install torch torchvision torchaudio

File Structure

All these files must be in same folder

index.py                # main program file
rnn.py                  # NEW: RNN classification and training system
ai_prompt.txt           # AI prompt template
credentials.json        # Google OAuth 2.0 credentials (from Google Cloud)
drive_token.pickle      # Auto-generated authentication token
rnn_models/             # NEW: Auto-created directory for model persistence
    ├── department_classifier_rnn.pth
    ├── vocab.pth
    ├── training_data.json
    ├── model_info.json
    └── performance_history.json

🗂️ Required Google Drive Structure

For optimal performance, organize your Google Drive as follows

Google Drive/
└── Company Reports/
    ├── finance/
    │   ├── Week-1.docx
    │   ├── Week-2.pdf
    │   └── Week-3.txt
    ├── marketing/
    │   ├── Week-1.gdoc
    │   └── Campaign-Report.txt
    └── IT/
        ├── Week-1.docx
        └── System-Status.pdf

📝 Technical Notes

GPU/CPU Compatibility

Important: The current implementation assumes GPU availability for PyTorch operations. If you're running on CPU-only systems, you'll need to modify the model loading in rnn.py:

# Change this line in setup_rnn_model():
self.rnn_model = torch.load(paths['model'], map_location='cpu', weights_only=False)

The code currently uses torch.load() without map_location='cpu' for optimal GPU performance.

💌 Possible Next Features Updates

These following are ideas for features that could be added in future versions.

Enhanced Training Data & MongoDB Integration

• Expended Initial Training Data
• MongoDB Atlas Integration
• Custom LSTM Architecture