Skip to main content
1536
Back to AI Lab

Embedding Visualizer

Watch AI understand semantic meaning in real-time

OpenAI EmbeddingsPCA1536D → 2D72 Points

What is this?

This visualization shows 72 portfolio items mapped in semantic space. Each dot represents work experience, projects, skills, or education—converted to 1536-dimensional vectors using OpenAI text-embedding-3-small, then reduced to 2D using PCA (Principal Component Analysis).

How it works: Text → 1536D vectors (OpenAI) → PCA → 2D visualization

→ Try it: Enter any text below to see how it maps against my experience.

This is how RAG (Retrieval-Augmented Generation) systems work under the hood!

How to read:

Points closer together are semantically similar. Note that PCA preserves ~15-30% of the original variance—balancing visualization clarity with implementation simplicity.

See this in action:

Visit the RAG Playground to compare how different retrieval strategies (Basic, Hybrid, Re-ranking) perform on this exact knowledge base! Watch real-time queries use these semantic clusters to retrieve the most relevant documents.

Generating embedding visualization...

Converting >20 knowledge base items into semantic space

Product Evolution

MVP → Validate → Improve → Know when to stop

Week 1🚀

MVP

Simplified Projection

First 2 dimensions only. Shipped in 1 day to validate concept.

Validated concept
Only 0.13% variance
✓ Shipped
Week 2

Upgrade

JavaScript PCA

Proper dimensionality reduction using PCA via ml-pca library.

15-30% variance
100x improvement
✓ Deployed
Not Built

Python UMAP

Best-in-class

UMAP offers better clustering, but adds ~1 week + pipeline complexity.

Marginal improvement
High complexity cost
✗ Stopped
💡

Product Manager Lesson

Ship fast to validate → Improve when you understand the problem deeply → Stop before over-engineering

PCA preserves ~17% variance vs UMAP's ~25%. Good enough for portfolio demo.
80% of the value, 20% of the complexity.