Algolia-style Postgres search with a Jev reranker

# postgres-search

Algolia-style search in plain Postgres: whole words, partial words, typos, barcodes, typeahead,
and facet counts, all in SQL. An optional step asks [Jev](https://docs.typesafe.ai) to push
clearly wrong results down the page.

The demo searches the USDA FoodData Central branded foods list. On the full 2025-12-18 release
(440,302 products), the warm results query takes about 2 ms for `cheerios`, 22 ms for `milk` and
53 ms for the misspelled `cheerois`; facet counts for `milk` take 128 to 135 ms
([measurements](docs/measurements.md)).

## Try it

Needs Docker and Node 22.18 or later.

```sh
git clone https://github.com/alexforman1/postgres-search
cd postgres-search
npm install
npm run db      # Postgres 16 in Docker on port 5432
npm run load    # 100,000-product sample, about 10 seconds
npm start       # http://localhost:3000
```

The server listens on `PORT` (default 3000). The server and scripts connect to `DATABASE_URL`
(default `postgres://postgres:postgres@localhost:5432/search_demo`). If port 5432 is taken, run the
`docker run` line from the `db` script in `package.json` with another host port, such as
`-p 127.0.0.1:5433:5432`, and point `DATABASE_URL` at it. The loader drops and recreates its
tables, so it refuses a database not named `search_demo` unless given `--any-database`. After a
reboot, start the database again with `docker start postgres-search`.

`npm run load -- --full` downloads the whole release (447 MB) instead of using the sample. It
needs `unzip`.

To try the Jev step, set a TypeSafe API key before `npm start`:

```sh
TYPESAFE_API_KEY=... npm start
```

Without a key everything else works the same. Jev's effect on the demo data has not been measured;
see [the Jev step](docs/jev.md).

## Use it with your data

Write one view, `searc