Users need an affordable, managed PostgreSQL option, ideally at a 'coffee price,' that includes the necessary stack (e.g., pgvector, integration with embedding generation) to easily build AI-powered semantic search for personal side projects and MVPs, simplifying infrastructure and reducing cost.
You don't need a specialized Vector Database to build AI apps. I have written ~170 newsletters over the last few years. The problem? Finding anything specific is a nightmare. Keyword search is brittle. If I search for "handling failures," I might miss an article about "exceptions" or "retries" because the words don't match exactly. I wanted ๐ฆ๐ฒ๐บ๐ฎ๐ป๐๐ถ๐ฐ ๐ฆ๐ฒ๐ฎ๐ฟ๐ฐ๐ต. It's the ability to search by concept rather than keywords. Usually, this means spinning up a specialized Vector Database. But for a side project or MVP, that often feels like overkill (and over-budget). So, I built it with ๐ฃ๐ผ๐๐๐ด๐ฟ๐ฒ๐ฆ๐ค๐ instead. Here is the "stack" I used: 1. .๐ก๐๐ง ๐ญ๐ฌ for the API. 2. ๐ข๐ฝ๐ฒ๐ป๐๐ to generate vector embeddings for my content. 3. ๐๐ถ๐๐ฒ๐ป ๐ณ๐ผ๐ฟ ๐ฃ๐ผ๐๐๐ด๐ฟ๐ฒ๐ฆ๐ค๐ to store the data and vectors. I enabled the pgvector extension, indexed my content, and now I can query "nearest neighbors" with simple SQL. The best part? I'm running this on ๐๐ถ๐๐ฒ๐ป'๐ ๐ป๐ฒ๐ ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐ฒ๐ฟ ๐ง๐ถ๐ฒ๐ฟ. For just $๐ฑ/๐บ๐ผ๐ป๐๐ต, I got a dedicated environment that is actually production-ready. - 8GB Storage (plenty for millions of embeddings) - Always-on (no "cold starts" or sleeping databases like other providers) - Managed (automated backups and updates included) It's perfect for the indie hacking or prototyping phase, where you want enterprise-grade tools without the enterprise price tag. You can spin up the exact same stack I used right here: ๐ https://fandf.co/4oC21k9 What would you build if you had a managed Postgres instance for the price of a coffee? P.S. Huge thanks to Aiven for sponsoring this post.