Sales Conversation Analyzer

AI-powered sales conversation analysis using your PremAI fine-tuned models.

Model Training

1. Visit studio.premai.io
2. Create dataset from /data/training_data_500_clean.jsonl (500 sample phone call transcripts)
3. Create snapshot
4. Fine-tune the model and see which one learns best
5. Run npm start and use UI to test the model

Setup

npm install
cp env.example .env
# Add your PREMAI_API_KEY to .env

Run

npm start

• Frontend: http://localhost:3066
• API: http://localhost:3033

Fine-Tuning Results

Conversation 1 - Input Transcript

Model 1 (Fine-tuned):

{
    "outcome_prediction": {
        "will_convert": true,
        "confidence": 0.85,
        "probability_score": 75
    }
}

Model 2 (Base):

{
    "outcome_prediction": {
        "will_convert": false,
        "confidence": 0.15,
        "probability_score": 23
    }
}

Full Fine-tuned Response | Full Base Response

Conversation 2 - Input Transcript

Model 1 (Fine-tuned):

{
    "outcome_prediction": {
        "will_convert": false,
        "confidence": 0.1,
        "probability_score": 20
    }
}

Model 2 (Base):

{
    "outcome_prediction": {
        "will_convert": false,
        "confidence": 0.25,
        "probability_score": 35
    }
}

Full Fine-tuned Response | Full Base Response

Conversation 3 - Input Transcript

Model 1 (Fine-tuned):

{
    "outcome_prediction": {
        "will_convert": false,
        "confidence": 0.35,
        "probability_score": 42
    }
}

Model 2 (Base):

{
    "outcome_prediction": {
        "will_convert": false,
        "confidence": 0.15,
        "probability_score": 25
    }
}

Full Fine-tuned Response | Full Base Response

Conversation 4 - Input Transcript

Model 1 (Fine-tuned):

{
    "outcome_prediction": {
        "will_convert": true,
        "confidence": 0.85,
        "probability_score": 75
    }
}

Model 2 (Base):

{
    "outcome_prediction": {
        "will_convert": true,
        "confidence": 0.85,
        "probability_score": 75
    }
}

Full Fine-tuned Response | Full Base Response

Structure:

• src/ - React TypeScript components
• api/ - Express route handlers
• data/ - Training datasets and transcripts
  • training_data_500_clean.jsonl - Training dataset
  • *.txt - Sample conversation transcripts
  • responses/ - Individual model response files
• compare.json - Model comparison analysis results (original)
• public/ - Static files