AI-Driven CRM Analytics System

A full-stack AI-powered CRM with sales tracking, event management, employee performance prediction, rewards, a PyTorch-based conversational assistant, and lead-classification ML - all in one web application. 82% accuracy lead scoring on 10,000+ customer records.

---

🎯 Problem

Small and mid-sized sales teams juggle CRMs, spreadsheets, calendar apps, HR tools, and analytics dashboards as separate products. Context gets lost between them, leads go cold, and employee performance is judged by gut feel instead of data.

💡 Solution

A single web application that brings everything under one roof:

• Customer & sales management - add, track, and segment customers
• Events calendar - schedule follow-ups, demos, team events
• Employee performance prediction - ML model forecasts rep performance
• Employee rewards - data-driven recognition tied to predicted performance
• Lead classification - KNN + Feedforward Neural Net scoring (82% accuracy)
• Conversational assistant - intent-based PyTorch chatbot for natural-language queries

📊 Impact / Results

• 82% accuracy on lead classification (KNN + Feedforward Neural Net)
• 10,000+ customer and sales records modeled in MySQL
• 6 integrated modules replacing what would otherwise be 4+ separate tools
• End-to-end: raw input → MySQL → ML features → web UI → conversational access

---

🏗️ System Architecture

---

🖥️ Application Walkthrough

Adding sales records

Events calendar

User profiles

Employee performance prediction

ML-driven forecasts of rep performance surfaced directly in the admin view - so managers review reps based on model output, not anecdote.

Employee rewards

Recognition tied back to predicted performance and achieved KPIs.

Conversational assistant

A PyTorch-trained intent classifier (bag-of-words → feedforward net, defined in model.py, trained via train.py on intents.json) lets users ask the CRM questions in natural language.

---

📈 Analytics & Insights

Trend and segmentation analyses rendered on the dashboard.

---

🛠️ Tech Stack

Layer	Tools
Language	Python 3.10+
Database	MySQL
Web framework	Flask, Jinja2
Data	Pandas, NumPy
ML - lead classification	scikit-learn (KNN + MLP / Feedforward NN)
ML - chatbot	PyTorch (intent classifier over intents.json)
Frontend	HTML, CSS, JavaScript

🚀 Getting Started

# Clone
git clone https://github.com/Isha2605/AI-Driven-CRM-Analytics-System.git
cd AI-Driven-CRM-Analytics-System/source_code

# Install Python dependencies
pip install -r requirements.txt

# (Optional) retrain models
python train.py       # trains the chatbot intent classifier → data.pth
python model.py       # trains the lead-classification models

# Run the Flask app
python flaskdemo1.py

Then open http://localhost:5000.

📁 Project Structure

├── assets/                 # Screenshots & architecture diagram (for README)
├── source_code/
│   ├── static/             # CSS, JS, images
│   ├── templates/          # Jinja2 HTML templates
│   ├── flaskdemo1.py       # Flask entrypoint
│   ├── form1.py            # Form handling
│   ├── implement1.py       # Business logic
│   ├── chatbot.py          # Chatbot runtime
│   ├── model.py            # ML model definitions
│   ├── train.py            # Training script
│   ├── intents.json        # Chatbot intent patterns
│   ├── data.pth            # Trained chatbot weights (PyTorch)
│   ├── sample.csv          # Sample dataset
│   └── bot.js              # Frontend chat widget
└── README.md

🔮 Future Work

• Swap KNN for Gradient-Boosted Trees (XGBoost) for production
• Add SHAP explanations alongside predictions in the UI
• Deploy on AWS (EC2 + RDS) with Nginx + Gunicorn
• Upgrade chatbot from intent-based to LLM + function calling over the CRM schema
• Role-based access control (RBAC) for admin vs rep views

📬 Contact

Isha Narkhede · Portfolio · LinkedIn · ishajayant207@gmail.com

📝 License

MIT - see LICENSE.