NovaChat 🚀

Multimodal AI Agent for Enterprise Conversations

NovaChat is an intelligent conversation analysis platform that transforms enterprise call transcripts into actionable insights through AI-powered analysis, interactive visualizations, and comprehensive reporting.

🌟 Features

🎯 Intelligent Query Routing

• Chain-of-Thought Reasoning: Advanced LLM router that understands context and conversation history
• Multi-Tool Selection: Automatically chooses the best analysis tool based on query intent
• Context Awareness: References previous responses and conversation flow

📊 Advanced Analytics & Visualizations

• Speaker Activity Analysis: Track speaking time, turn counts, and engagement patterns
• Sentiment Trend Analysis: Monitor emotional progression throughout conversations
• Topic Categorization: Automatically categorize conversation segments by themes
• Custom Chart Generation: Create bar charts, pie charts, and line graphs on demand

📄 Comprehensive Reporting

• PDF Report Generation: Professional downloadable reports with detailed analysis
• Audio Summaries: Text-to-speech narration of key insights
• Interactive Dashboards: Real-time conversation profiling and metrics

🔊 Audio Processing

• Multi-format Support: Process MP3, WAV, M4A audio files
• Automatic Transcription: Convert speech to text with speaker identification
• Real-time Analysis: Instant insights from uploaded conversations

🏗️ Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   Streamlit UI  │───▶│  Query Router    │───▶│  Analysis Tools │
│                 │    │ (Chain-of-Thought)│    │                 │
└─────────────────┘    └──────────────────┘    └─────────────────┘
         │                       │                       │
         ▼                       ▼                       ▼
┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│ Audio Processor │    │   ChromaDB       │    │  Google Gemini  │
│                 │    │   Vector Store   │    │      LLM        │
└─────────────────┘    └──────────────────┘    └─────────────────┘

🚀 Quick Start

Prerequisites

• Python 3.8+
• Google Gemini API Key
• ffmpeg (for audio processing)

Installation

1. Clone the repository

   git clone https://github.com/yourusername/NovaChat.git
   cd NovaChat

2. Install dependencies

   pip install -r requirements.txt

3. Set up environment variables

   # Create .env file
   echo "GOOGLE_API_KEY=your_gemini_api_key_here" > .env

4. Run the application

   streamlit run app.py

5. Open in browser Navigate to http://localhost:8501

📁 Project Structure

NovaChat/
├── app.py                      # Main Streamlit application
├── query_engine.py            # Intelligent query routing system
├── outputs.py                 # Analysis tools and LLM integration
├── pdf_builder.py             # PDF report generation
├── normalization.py           # Audio processing and transcription
├── ingest.py                  # Vector database ingestion
├── profiler.py               # Conversation profiling
├── transcript_parser.py       # JSON transcript parsing
├── config.py                 # Configuration settings
├── requirements.txt          # Python dependencies
├── .env                      # Environment variables
├── chroma_db/               # Vector database storage
├── temp_uploads/            # Temporary file storage
└── README.md                # This file

🛠️ Core Components

Query Router (query_engine.py)

Smart routing system that uses chain-of-thought reasoning to select the best analysis tool:

# Example: Intelligent routing based on query intent
query = "Show me how Michelle responded to fee discussions"
# Router analyzes: intent → context → tool selection → execution

Analysis Tools (outputs.py)

Specialized functions for different types of analysis:

• build_holistic_analysis_chart() - Complex multi-dimensional analysis
• build_speaker_activity_chart() - Speaker engagement metrics
• build_sentiment_trend_chart() - Emotional progression tracking
• build_pdf_response() - Comprehensive report generation

Audio Processing (normalization.py)

Handles audio file conversion and transcription:

# Automatic transcription with speaker identification
transcribe_audio("conversation.mp3") → structured_transcript.json

💡 Usage Examples

Basic Conversation Analysis

Upload audio → "Summarize the main discussion points"

Advanced Analytics

"Create a breakdown chart showing time spent on fees vs. program details"
"Analyze Michelle's engagement level throughout the call"
"Generate a sentiment trend for Speaker0"

Report Generation

"Generate a comprehensive report analyzing Michelle's concerns and responses"
"Create a PDF summary of this sales conversation"

Speaker Analysis

"Who talked the most and for how long?"
"Show me the number of times each speaker contributed"
"Compare speaking patterns between participants"

🔧 Configuration

Environment Variables (.env)

GOOGLE_API_KEY=your_gemini_api_key_here
CHROMA_DB_PATH=./chroma_db
COLLECTION_NAME=conversation_segments
EMBED_MODEL_NAME=all-MiniLM-L6-v2

Model Configuration (config.py)

MODEL_ID = "gemini-2.5-flash"  # Google Gemini model
EMBED_MODEL_NAME = "all-MiniLM-L6-v2"  # Sentence transformer
CHROMA_DB_PATH = "./chroma_db"

📊 Supported File Formats

Audio Files

• MP3, WAV, M4A, FLAC
• Automatic conversion to WAV for processing

Transcript Files

• JSON format with speaker identification
• Structured conversation segments with timestamps

Output Formats

• Interactive charts (matplotlib)
• PDF reports (comprehensive analysis)
• MP3 audio summaries (text-to-speech)

🤖 AI Models & Integration

• Google Gemini 2.5 Flash: Primary LLM for analysis and reasoning
• Sentence Transformers: Vector embeddings for semantic search
• CardiffNLP RoBERTa: Sentiment analysis pipeline
• Google Text-to-Speech: Audio summary generation

🎯 Key Features in Detail

Chain-of-Thought Routing

Step 1: Analyze query intent (chart vs. report vs. summary)
Step 2: Consider conversation history and context
Step 3: Reference previous responses if applicable
Step 4: Select optimal tool for task
Step 5: Execute with full context awareness

Conversation Profiling

• Automatic detection of conversation type (Sales, Support, etc.)
• Key entity extraction (names, products, companies)
• Context preservation across queries

Vector-Based Retrieval

• ChromaDB integration for semantic search
• Contextual segment retrieval
• Relevant information extraction for queries

🚨 Troubleshooting

Common Issues

1. API Key Errors

   # Verify .env file exists and contains valid API key
   cat .env

2. Audio Processing Fails

   # Install ffmpeg
   # Ubuntu/Debian: sudo apt install ffmpeg
   # macOS: brew install ffmpeg
   # Windows: Download from ffmpeg.org

3. Dependencies Missing

   pip install --upgrade -r requirements.txt

🖥️ Presentation & Demo

• Project Presentation (PPT): View Google Slides
• Demo Video: Watch Demo

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Google Gemini for advanced language model capabilities
• ChromaDB for vector database functionality
• Streamlit for the interactive web interface
• HuggingFace for pre-trained models and pipelines

---

Built with ❤️ for enterprise conversation intelligence

For questions or support, please open an issue or contact the development team.