I made SEO tool using vector embeddings

Hacker News

Share card

I made SEO tool using vector embeddings

Hey HN, I’m a solopreneur and have been in the SaaS space for the past 3 years. My last three startups failed, and a common challenge in all of them was figuring out how to reach my audience. SEO always seemed like the answer, but I didn’t know where to start. It felt overwhelming—so many technical terms like keyword research, clustering, SERP, semantics, topical authority… I could go on. That’s where my idea for my next startup came from: building an all-in-one SEO tool that’s easy to use and understand. The complexity of SEO is hidden behind a simple UI, and the only thing users need to do is publish the generated articles. After working on it for the past 4 months, I just launched and already have a few paying customers through Reddit! One of the biggest technical challenges was figuring out how to prioritize what to write about. Every customer is in a different niche, with a unique audience and offering. At first, I tried using LLMs to filter and prioritize topics, but it didn’t work well. Many irrelevant topics slipped through, and customers weren’t happy. Then, I came across an article about topical authority and vector embeddings. That was my breakthrough! Here’s what I did: •I gathered all the keywords a customer’s website already ranks for. •I created vector embeddings for those keywords. •I built a function that uses cosine vector distance to measure the similarity between a new article topic and the site’s existing ranked keywords. It works like a charm! This method helps me prioritize articles related to a website’s core offering first, then expand into supporting (pillar) topics. I assign each topic a score from 1 to 100 based on its relevance. Next, I plan to use embeddings for internal linking, categorization, recommended reads, and more. There’s still a lot to do, but I’m excited about where this is going. I’ve learned so much in the past three weeks—let’s see where it takes me! Would greatly appreciate if you guys could try it out and provide feedback on the articles!