Embedding Explorer — See Your Text in 2D

This is a simple, interactive Streamlit app I built to make text embeddings less mysterious and a lot more fun to explore. You can:

• Upload a CSV with a text column (or just paste your text directly)
• Pick an embedding model from sentence-transformers
• Reduce the vectors to 2D with UMAP or t-SNE
• Color-code them with KMeans clusters (optional)
• Zoom, hover, filter, and download the results
• No API keys, no setup headaches — everything runs locally.

Try It Out

Live app: [https://zasaee-text-embedding-visualizer-app-qmmew9.streamlit.app/]

Want a quick example? Use the included sample_texts_multitheme.csv. It’s got four themes (ML, Healthcare, Finance, Manufacturing) so you’ll see nice, clean clusters right away.

Why I Built This?!

• Embeddings are everywhere in modern ML and LLM workflows, but they’re hard to “see.” I wanted a quick way to:
• Explain embeddings visually when talking to non-technical folks
• Explore my own datasets before training a model
• Share an easy, no-login demo link with anyone
• This project also shows the kind of ML tooling I like to build: lightweight, clear, and ready to use.

Example Run

Step 1 – Upload and set parameters

Step 2 – Visualize clusters in 2D

Example Output

The Download embeddings CSV button gives you a file like this:

text	cluster	x	y	emb_0	emb_1	...
"Transformers are powerful models..."	0	-5.12	2.33	0.018	-0.045	...
"AI assists early cancer detection..."	1	3.41	-1.87	-0.012	0.032	...

How to Run Locally

git clone https://github.com//embedding-explorer.git cd embedding-explorer pip install -r requirements.txt streamlit run app.py

Deploy It Yourself (Free)

• Streamlit Community Cloud: Connect your GitHub repo → pick app.py → deploy.
• Hugging Face Spaces: Create a new Space → choose Streamlit → point it to your repo.

What’s Happening Under the Hood

• Embeddings: sentence-transformers models map your text to high-dimensional vectors.
• Dimensionality Reduction: UMAP/t-SNE compress those vectors into 2D so you can see patterns.
• Clustering: KMeans assigns points to groups (optional).
• Filtering: Quickly find and highlight points containing a specific word/phrase.