Using Embeddings to Connect Meaning

One of the biggest challenges in professional communities? Helping people find the right connections and conversations.

At Guild, we had thousands of communities, each fostering unique discussions. Initially, we assumed increasing engagement meant helping members join more communities and connect with more people. Our first idea? Traditional tagging—categorizing communities and members by industry, discipline, and interests.

But we quickly hit limitations:

❌ Tags are rigid and require manual input

❌ They don’t capture the real nuances of conversations

❌ They reinforce existing silos rather than revealing unexpected, valuable connections

So we took a different approach: AI embeddings.

What’s an embedding?

An embedding is a way of representing text (or any data) as a numerical vector in multi-dimensional space. Essentially, it captures meaning and context—so that words, phrases, or even entire conversations that have similar themes end up closer together.

Instead of tagging communities with fixed labels, we used OpenAI’s embedding function to turn both community discussions and individual conversations into numerical vectors. This allowed us to calculate relatedness:

🔹 How similar are two people’s conversations?

🔹 How closely does a community’s content align with an individual’s discussions?

Rather than relying on self-reported interests or rigid categories, we created a dynamic map of interests, passions, skills, and knowledge across traditional boundaries.

The result? Phenomenal connections—people discovering communities and collaborators they never would have found otherwise. And it worked: user engagement grew by 12%.

This method has huge potential beyond professional networking. Imagine using AI-driven embeddings to break down silos in hiring, learning, or even innovation.