gstarwd / yc-idea-matcher

Submit your idea and get a list of similar ideas that YCombinator has invested in in the past.

YC idea matcher

This project allows you Submit your idea and get a list of similar ideas that YCombinator has invested in before.

The project is built using the following technologies:

• Neon: Serverless Postgres
• pg_embedding: open-source Postgres extension for vector storage and similarity search
• Neon Serverless Driver
• Next.js: Fullstack framework for React
• Vercel: deployment platform
• OpenAI API: generating vector embeddings
• TailwindCSS: Utility-first CSS framework
• Upstash Redis: serverless Redis for rate limiting
• Zod: TypeScript-first schema validation
• React Query: data fetching and caching library
• Vaul: Drawer component for React.

How the app works

You will find a script called generate-embeddings.ts located in the root directory of this project. After running npm run generate-embeddings, the script does the following:

1. It creates the database schema and installs the pg_embedding extension
2. It goes through the YCombinator API 'https://api.ycombinator.com/v0.1/companies?page=1' and gets all the companies
3. For each company it generates embeddings using the long description and then stores the company data in the database.

Some companies don't have a long description, so we needed to manually remove those from the database by running delete from companies WHERE embedding = ARRAY[]::real[];

The app itself is a Next.js app with an API route located at /api/idea. Whenever a user submits an idea, the following happens:

1. The idea is sent to the OpenAI API to generate an embedding
2. We then use pg_embedding to retrieve the top 3 most similar ideas