How Search Works

Overview

Search on Liners uses a hybrid approach — we run traditional keyword matching and AI-powered semantic search at the same time, then merge the results. The idea is simple: keywords are great when you know exactly what you're looking for, but semantic and natural language search catches the products you'd miss otherwise. QA Quinn keeps an eye on search quality to make sure results stay relevant.

How It Works

When you type a query, two things happen in parallel:

1. Keyword search — We match your query against product names, taglines, and descriptions. Fast, reliable, always available.
2. Semantic search — Your query gets converted into a vector embedding (a numerical fingerprint of its meaning), and we compare it against every product's embedding to find semantically similar matches.

Results from both are merged and ranked:

• Exact name matches always float to the top.
• Semantic matches are ranked by how similar they are to your query.
• Keyword-only matches fill the remaining slots.
• Duplicates are automatically removed.

Example

Searching for "send money abroad" will surface products related to international transfers and remittances — even if they don't contain those exact words anywhere in their listing.

How Products Are Embedded

Each product is turned into a numerical vector that captures what it does, who it's for, and where it operates. The embedding is built from:

• Product name and tagline
• Short description
• Long description
• Categories and tags
• Countries

We use Google text-embedding-005 (768 dimensions) for generating these embeddings. When a new product is added, it gets embedded automatically. When a product is updated, it trigger a re-embed.

Search Quotas

Semantic search costs money (each query hits an embedding API), so we have daily usage limits to keep things sustainable:

When you've used up your quota:

• Search automatically falls back to keyword-only results — still fully functional, just without the semantic layer.
• You'll see a message letting you know, with a suggestion to sign in (for guests) or that you've hit your daily cap (for logged-in users).
• Your quota resets every 24 hours.

To be fair, I see no reason why anyone should want to search more than 50 times in 24 hours, but I know humans can be unpredictable.

Fallback Behaviour

If the embedding API is ever unavailable (timeout, error, or anything unexpected), search gracefully falls back to keyword-only results. You won't see an error — it just works, minus the semantic layer. LGTM Larry monitors for these fallbacks to make sure they're rare.