chaas/debshots
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages 1 Wiki Activity Actions

Add semantic package search via pgvector embeddings

Browse source 
Package text search now combines several strategies: exact name match,
name prefix promotion, compound word splitting ("sqlite browser" finds
"sqlitebrowser") and semantic nearest neighbor search over description
embeddings computed by an external embedding service (all-MiniLM-L6-v2,
384 dims) stored with the pgvector extension. Classic PostgreSQL
full-text search remains as fallback when the vector service is
unreachable. Gibberish queries without lexical overlap with the package
data return empty results instead of random matches.

Embeddings can be backfilled with bin/rails debshots:compute_vectors.

Also drops the unused lograge gem from the Gemfile.
This commit is contained in:
Christoph Haas 2026-08-22 22:07:56 +02:00
parent c7e8109bbd
commit 4948fe50e3
15 changed files with 662 additions and 125 deletions
Download patch file Download diff file Expand all files Collapse all files

24
README.md
View file

@ -15,6 +15,30 @@ https://salsa.debian.org/debian/debshots/-/issues
Check out the doc/README.Development.md
## Package search
The package search combines several strategies, tried in this order:
1. **Exact package name**: Searching for `vim` shows exactly this
package without calling any external service.
2. **Name prefix matches**: Packages whose name starts with the query
rank first, e.g. searching for `sqlite` promotes `sqlite3`,
`sqlitebrowser` and friends above all other results.
3. **Compound word splitting**: Packages whose name contains *all*
query tokens match too, so `sqlite browser` finds `sqlitebrowser`.
4. **Semantic search**: The query is embedded via the vector service
(all-MiniLM-L6-v2, 384 dimensions) and the nearest neighbors among
the package description embeddings are returned, using the pgvector
extension with an HNSW cosine index. Embeddings are computed from
the package description and long description and can be (re-)built
with `bin/rails debshots:compute_vectors`.
5. **Full-text fallback**: If the vector service cannot be reached,
classic PostgreSQL full-text search takes over.
Queries that share no vocabulary with the package data (gibberish,
keyboard mash) return an honest empty result instead of random
semantic matches.
## Deployment
Read the doc/README.Installation.md