📊 Mini Embedding Explorer

A lightweight, business‑ready analytics tool that transforms raw text into visual insights.
Upload customer feedback, support tickets, policies, CVs, or any text dataset — the app generates embeddings, reduces them to 2D, clusters them, and reveals hidden patterns.

Perfect for SMEs, councils, charities, and teams that need fast, AI‑powered text understanding without complex infrastructure.

---

📸 Screenshots

• Dashboard and CSV preview
• Scatter graph of embeddings
• Cluster summary of the CSV data

---

🚀 Features

• Upload CSV or TXT files
• Choose an embedding model (MiniLM by default)
• Automatic dimensionality reduction (PCA or UMAP)
• KMeans clustering with adjustable cluster count
• Interactive 2D scatter plot
• Hover to view original text
• Download clustered results as CSV
• Zero configuration — runs locally or on Streamlit Cloud

---

🧠 Real‑World Use Cases

This tool is intentionally small but delivers real business value:

• Customer Feedback Clustering — identify themes in reviews and surveys
• Support Ticket Triage — group recurring issues and detect outliers
• Document Similarity — map policies, reports, and internal knowledge
• Fraud Pattern Discovery — spot repeated scam wording or anomalies
• HR & Recruitment Insights — compare CVs and job descriptions
• SEO Topic Mapping — visualise content clusters and gaps

---

📦 Installation

pip install -r requirements.txt

▶️ Run the App

streamlit run app.py

---

📁 Project Structure

mini-embedding-explorer/
│
├── app.py                 # Streamlit UI
├── screenshots/
├── embeddings.py          # Embedding + clustering engine
├── requirements.txt       # Dependencies
├── README.md              # Project documentation
└── examples/
    └── feedback.csv

---

🧩 How It Works

• Embeddings
  Text is converted into numerical vectors using a SentenceTransformer model.

• Dimensionality Reduction
  High‑dimensional vectors are compressed into 2D using PCA or UMAP.

• Clustering
  KMeans groups similar texts together.

• Visualisation
  The 2D points are plotted so humans can see patterns instantly.

---

📊 Example Workflow

• Upload a CSV of customer comments

• Select the text column

• Choose PCA or UMAP

• Pick number of clusters

• Generate embeddings

• Explore clusters visually

• Download results

---

📝 Example Use Case: Customer Feedback

A business uploads:

feedback.csv

• "Delivery was late again"
• "Website login keeps failing"
• "Customer service was excellent"
• "Refund process is confusing"

The tool reveals clusters like:

• Delivery issues

• Website bugs

• Positive service comments

• Refund complaints

• This helps teams prioritise improvements.

---

🛣️ Future Roadmap

• Advanced Embedding Models — Add support for larger or domain‑specific models (legal, financial, medical) to improve clustering accuracy for specialised industries.

• Semantic Search Engine — Allow users to search their dataset using natural language queries powered by embeddings, turning the tool into a mini knowledge explorer.

• Topic Labeling — Automatically assign human‑readable labels to clusters (e.g., “Delivery Issues”, “Refund Complaints”), making insights easier for non‑technical teams.

• Interactive Cluster Editing — Let users merge, rename, or split clusters directly in the UI, enabling custom business workflows and cleaner reporting.

---

• Built by Roy Peters 😁