Softpro : Customer Analytics System – Sentiment & Sales Insights

AI-powered Audio Transcription, Sentiment Analysis & Sales Insights Platform

---

🚀 Live App:

Table of Contents

• Overview
• Features
• System Workflow
• Tech Stack
• Project Structure
• Training Output (PDF Images)
• How to Run
• Dataset Requirements
• Sentiment Model Modes
• Analytics Provided
• Recommendation Engine
• Future Scope
• Author
• License

---

#Project Report:

click here go to report and Download [Final_report.docx] ("https://github.com/Ramesh8dsaiml/Customer-Analytics-System-Sentiment-Sales-Insights/blob/main/Final_report.docx"

Overview

softpro-analytics

(company_name - softpro)

#AI-powered platform for audio transcription, sentiment analysis, CRM merging & sales insights.

Softpro Analytics – Sentiment & Sales Insights

A complete AI-powered Streamlit platform that converts audio counselling calls + CRM logs into:

• Structured transcripts
• Sentiment insights
• Negative keyword patterns
• Tech-stack & location-based analytics
• Actionable recommendations to improve conversions
• Built using Whisper/Vosk ASR, Transformers, Scikit-Learn, and an interactive Streamlit dashboard.

------------------------------------------------------------------->

Features

Audio Processing

• Upload MP3/WAV call recordings
• High-quality speech-to-text using Whisper (OpenAI)
• Offline support via Vosk

CRM Log Processing

• Upload CSV logs
• Auto-map columns
• Merge call transcripts + counselor remarks

Sentiment Analysis

• Pretrained DistilBERT (Binary: pos/neg)
• OR custom ML model (TF-IDF + Logistic Regression)

✔ Interactive Analytics Dashboard

• Sentiment distribution
• Location-wise analysis
• Tech-stack-wise performance
• Monthly sentiment trend
• Top negative keywords

✔ Recommendation Engine

Automatically identifies issues:

• Fees
• Timing
• Location
• Placement
• Faculty support

And generates actionable suggestions.

✔ Export final processed dataset

------------------------------------------------------------------------------------------------------------------>

System Workflow

Audio Files ↓ Transcription (Whisper/Vosk) ↓ CRM Merge ↓ Sentiment Analysis ↓ Analytics Dashboard ↓ Recommendations

---

🛠️ Tech Stack

Languages & Libraries

• Python 3.10+
• Streamlit
• Whisper / Vosk
• Hugging Face Transformers
• Scikit-Learn
• Pandas, NumPy
• Plotly

AI Models

• Whisper ASR (tiny/base/small/medium)
• DistilBERT Sentiment model
• Logistic Regression (Custom training option)

---

📂 Project Structure

softpro-analytics/ │── app.py # Main Streamlit App │── requirements.txt # Package list │── sample.csv # Demo CRM Log │── recordings/ # Demo audio (optional) │── README.md # Project documentation │── softpro_page_1.png │── softpro_page_2.png │── softpro_page_3.png │── screenshots/ # Dashboard images

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------@

Training Output (From PDF)

page 1

### page 2 ### Page 3 ### Page 4 ### Page 5

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------@

How to Run Locally

1. Create Virtual Environment

python -m venv myenv
myenv\Scripts\activate     # Windows


. Install Dependencies
pip install -r requirements.txt

If Whisper gives FFmpeg error:
Add this to system PATH:


#Run App
streamlit run app.py

-----> Dataset Requirements
CRM Log CSV (Required Columns)
student_name  
year  
tech_stack  
location  
remarks  
call_id  
date  
label (optional: positive/neutral/negative)

----> Audio Support

mp3, wav, m4a, aac

Whisper auto-converts

Vosk requires WAV, 16-bit PCM, mono

-------> Sentiment Model Modes
🔹 Pretrained Mode (Default)

Uses Huggingface DistilBERT

Outputs: positive / negative + confidence

🔹 Custom Training Mode

Triggered when CSV has a label column

Uses TF-IDF + Logistic Regression

Generates Classification Report



------> Analytics Provided

 Sentiment Distribution

 Location-wise Sentiment Comparison

 Tech-stack-wise Analysis

 Monthly Sentiment Trend

 Negative Keyword Extraction (TF-IDF)

 Keyword-based Objection Patterns



------> Recommendation Engine

Automatically detects issues and generates suggestions:

Issue Type	Recommended Action
Fees	EMI plans, scholarships, limited-time offers
Timing	Add evening/weekend batches
Placement	Highlight alumni success, workshops
Location	Provide hybrid/online options
Faculty/Support	Extra mentor hours, doubt sessions



----> Future Scope

Hindi/Hinglish ASR Support

Real-time CRM Integration

Emotional Tone Detection

Dynamic Lead Scoring

Mobile Responsive UI

Author

Ramesh Kumar
B.Tech AI & Data Science (2022–2026)
(AKTU Lucknow)
Future Institute of Engineering & Technology, Bareilly

License

This project is built for academic and educational purposes.