I Rebuilt the PlayPhrase.me Search Engine

Small update from the engine room.

For the last few days, I have been rebuilding one of the most important parts of PlayPhrase.me: the search engine.

Not the search box you see on the page. Not the buttons. The actual thing underneath - the part that has to search through 39.5 million phrases and still feel instant.

The old phrase search was already fast enough. Usually it took around 50-100 ms to find matching phrases, which is not bad at all. The problem was everything around it.

It needed about 120 GB of RAM.

It took around 20 seconds to start.

And, more importantly, it had become very hard to extend.

So when people asked for things that sounded simple - filter by year, genre, director, better Clip Search controls, more ways to explore the database - the answer was not really "I don't want to build it." The answer was closer to: the old engine did not leave enough room.

It worked, but it was too heavy.

So I rebuilt it from scratch.

The new version is written in Rust, using everything I have learned from running PlayPhrase.me for more than 10 years. It is not just the same search rewritten in another language. The internal approach is different: smaller runtime index, faster startup, more predictable memory use, and a structure that should be much easier to expand.

The numbers are a bit ridiculous:

Before:

- about 120 GB RAM

- around 20 seconds startup

- 50-100 ms for phrase search

After:

- about 2 GB RAM

- 68 ms startup

- under 1 ms for the actual phrase search

So the runtime index is roughly 60x smaller.

This is the kind of change that may not look dramatic from the outside on day one. There is no giant new button on the homepage saying "the internal search index is now sane." But for the project, this is a big milestone.

It removes a real limit.

The first place where I want to use this new room is Clip Search. I don't plan to suddenly put every advanced filter everywhere on the site. Clip Search is the natural place for it, because it is already for people who need more control: creators, editors, teachers, and anyone trying to find a very specific clip quickly.

So the direction is pretty clear: more filters, more control, better discovery, and better ways to work with the PlayPhrase.me database beyond just typing text into a box.

One small side note: AI has also become part of how I build now. I use it mostly for specs, architecture notes, code review, and turning spoken ideas into technical tasks. It helps me move faster, but the product decisions, constraints, testing, and final responsibility are still mine.

For PlayPhrase.me, this new search engine is one of those internal changes that makes future work much less painful.

Smaller runtime.

Faster startup.

Much more space to build on top of it.

And that means I can finally start adding some of the search and discovery features that were blocked for a long time.

One important note: this is not live on the site yet. Right now it is running only in my internal testing environment. If everything goes well, I plan to roll it out in about a week.