DeepMost Agentic SDR

AI-Powered Sales Development Representative Simulation System

🚀 Now with Premium v2.0 - Advanced Analytics & ML-Powered Insights!

This project uses Multi-Agent AI to simulate B2B sales conversations for training downstream models and analyzing sales strategies.

---

✨ What's New in v2.0

🧠 Advanced Analytics Engine

• Win Probability Prediction - ML model predicts deal success in real-time
• Sentiment Trajectory Analysis - Track buyer engagement throughout the call
• Objection Pattern Detection - Automatically categorize and analyze objections
• Feature Importance - Understand what drives successful outcomes

📊 Interactive Dashboards

• Plotly Visualizations - Beautiful, interactive charts
• Real-time Insights Panel - Live coaching during simulations
• Performance Trends - Track improvement over time
• Comprehensive Analytics - Gauge charts, sunbursts, and radar plots

🎯 Real-Time Coaching

• Live Suggestions - Get coaching tips during conversations
• Urgency Indicators - Know when to pivot your approach
• Win Probability Updates - See predictions update turn-by-turn

---

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Key

Create a .env file with your Gemini API key:

GEMINI_API_KEY="your-api-key-here"

3. Run the Application

# Interactive Web UI
python app.py

# Batch Processing Pipeline
python main.py

---

📊 Data Collection & Analysis

Generated Data Files

File	Location	Description
simulations_master.csv	data/processed/	One row per simulation with all metadata
conversation_turns.csv	data/processed/	One row per message for NLP analysis
simulation_metrics.csv	data/processed/	ML-ready features with binary outcomes
<uuid>.json	data/raw/conversations/	Full conversation with context

Data Schema

simulations_master.csv

Column	Type	Description
simulation_id	UUID	Unique identifier
timestamp	ISO8601	When simulation ran
target_url	String	Company website
num_turns	Int	Number of conversation turns
total_seller_words	Int	Total words by seller
total_buyer_words	Int	Total words by buyer
score	Int (1-10)	AI-assessed call quality
outcome	String	Success/Failure
key_objection	String	Price/Timing/Authority
feedback	String	AI coaching feedback

simulation_metrics.csv (ML-Ready)

Column	Type	Use Case
outcome_binary	0/1	Classification target
word_ratio_seller_buyer	Float	Feature engineering
total_conversation_length	Int	Regression feature

---

📈 EDA & Modeling Guide

Loading Data in Python

import pandas as pd
import json

# Load aggregated data
df = pd.read_csv('data/processed/simulations_master.csv')

# Load turn-level data for NLP
turns = pd.read_csv('data/processed/conversation_turns.csv')

# Load ML-ready metrics
metrics = pd.read_csv('data/processed/simulation_metrics.csv')

# Load individual conversation
with open('data/raw/conversations/<simulation_id>.json') as f:
    conv = json.load(f)

Sample EDA Queries

# Success rate by objection type
metrics.groupby('objection_type')['outcome_binary'].mean()  

# Average conversation length by outcome
metrics.groupby('outcome_label')['total_conversation_length'].mean()

# Word ratio analysis
metrics[metrics['outcome_binary']==1]['word_ratio_seller_buyer'].describe()

Sample ML Model

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# Features
X = metrics[['total_conversation_length', 'word_ratio_seller_buyer', 
             'seller_avg_words_per_turn', 'buyer_avg_words_per_turn']]
y = metrics['outcome_binary']

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

model = RandomForestClassifier(n_estimators=100)
model.fit(X_train, y_train)
print(f"Accuracy: {model.score(X_test, y_test):.2f}")

---

🏗️ Project Structure

DeepMost_Agentic_SDR/
├── app.py                      # Gradio Web UI
├── main.py                     # Batch processing pipeline
├── requirements.txt            # Dependencies
├── .env                        # API keys (not in git)
├── README.md                   # This file
├── src/
│   ├── __init__.py
│   ├── agent_logic.py          # Multi-agent simulation engine
│   ├── scraper.py              # Web scraping module
│   └── data_manager.py         # Data persistence layer
├── data/
│   ├── raw/
│   │   └── conversations/      # JSON conversation files
│   └── processed/
│       ├── simulations_master.csv
│       ├── conversation_turns.csv
│       └── simulation_metrics.csv
└── notebooks/                  # Jupyter notebooks for EDA
    └── eda_template.ipynb

---

🔬 Features

Multi-Agent System

• Seller Agent: Trained on value-based selling techniques
• Buyer Agent: Skeptical CTO persona with realistic objections
• Judge Agent: Analyzes call quality and provides coaching

Real-Time Streaming

• Watch conversations unfold in real-time in the web UI
• Turn-by-turn updates with status indicators

Comprehensive Analytics

• Outcome distribution charts
• Score histograms
• Objection type analysis
• Conversation length correlation

Data Science Ready

• Structured CSV exports for pandas
• JSON files for NLP/dialogue analysis
• ML-ready features with binary targets
• UUID tracking for reproducibility

---

📝 License

This project is for educational/internship purposes.

---

🤝 Contributing

Feel free to submit issues and enhancement requests!