CX‑360 Insight Engine

AI‑Powered Customer Feedback Analysis with RAG, Clustering, Sentiment & Business Insights

The CX‑360 Insight Engine is an end‑to‑end customer‑experience analytics system built using:

• LLM‑powered RAG (Groq + Llama 3.1)
• Semantic embeddings
• Topic clustering (t‑SNE + KMeans)
• Sentiment analysis
• Cluster naming using LLMs
• Interactive Streamlit dashboard
• Business‑ready insight generation

This project transforms raw customer feedback into actionable insights, root‑cause analysis, and prioritized recommendations — similar to enterprise VoC platforms like Medallia, Qualtrics, and Clarabridge.

Key Features

1. Intelligent RAG‑Based Insights

Ask any question about customer feedback and the engine will:

• Retrieve the most relevant comments
• Analyze sentiment distribution
• Summarize themes
• Provide executive‑ready insights
• Recommend business actions
• Generate a Customer Recovery Plan if negative sentiment is high

2. Topic Clustering (with Stable LLM‑Generated Names)

The system automatically:

• Embeds all feedback
• Reduces dimensionality (t‑SNE)
• Clusters comments into topics
• Uses an LLM to name each cluster
• Saves names in cluster_names.json so they never change
• Visualizes clusters in a 2D scatter plot

You also get:

• A dropdown to inspect each cluster
• All feedback belonging to that cluster
• Clear explanation of what X/Y axes mean

3. Sentiment Analysis

Each comment is labeled as:

• Positive
• Neutral
• Negative

Sentiment distribution is used in:

• Insight generation
• Cluster summaries
• Business recommendations

4. Business Insight Engine

For every query, the system generates:

• Executive summary
• What customers are saying
• Sentiment breakdown
• Root‑cause analysis
• Business impact
• Prioritized recommendations
• Customer recovery plan (if needed)
• Strengths (if sentiment is positive)

How It Works

1. Data Processing

• Clean text
• Compute embeddings
• Compute sentiment
• Store results

2. RAG Retrieval

• Retrieve top 300 semantically similar comments
• Compute sentiment distribution
• Pass context to LLM

3. Insight Generation

LLM produces:

• Themes
• Patterns
• Recommendations
• Recovery plan
• Executive summary

4. Topic Clustering

• t‑SNE → 2D coordinates
• KMeans → cluster labels
• LLM → cluster names
• JSON → persistent naming

Cluster Visualization

The scatter plot shows:

• Each point = one customer comment
• Colors = cluster groups
• X/Y axes = semantic dimensions (not sentiment)
• Negative/positive values = position in embedding space

Interpretation:

• Points close together → similar meaning
• Points far apart → different topics

Tech Stack

• Core
• Python
• Streamlit
• Groq LLMs (Llama‑3.1‑8B‑Instant / Llama‑3.3‑70B‑Versatile)
• SentenceTransformers
• scikit‑learn
• UMAP / t‑SNE
• Pandas
• LLM
• Groq API
• RAG retrieval
• Insight generation
• Cluster naming

Project Structure

Code CX-360/ │ ├── app.py ├── requirements.txt ├── README.md │ ├── src/ │ ├── rag.py │ ├── prompts.py │ ├── clustering.py │ ├── sentiment.py │ ├── utils.py │ └── cluster_names.json │ └── data/ └── sample_feedback.csv 🔧 Setup Instructions

1. Install dependencies Code pip install -r requirements.txt
2. Set your Groq API key Code export GROQ_API_KEY="your_key_here" (Linux/Mac syntax) set GROQ_API_KEY=your_key_here (Windows)
3. Run the app Code streamlit run app.py

Future Improvements

• Add topic‑level sentiment dashboards
• Add SHAP explanations for sentiment model
• Add multi‑language support
• Add PDF export for insights
• Add automated weekly CX reports

Why This Project Stands Out

This is not a basic NLP project — it is a full customer‑insight engine combining:

• Machine learning
• LLM reasoning
• RAG
• Clustering
• Visualization
• Business analytics