Regalos Mejores: Embedding-Powered Gift Search at Scale
A self-built Spanish gift-discovery engine that turns natural-language requests into curated Amazon picks using embedding search and an LLM enrichment pipeline.
01 The Problem
Keyword search doesn't understand a request like "regalo para mi madre, cumple 60 y le encanta la jardineria" - it just matches the literal words. Generic gift-guide sites solve this with static listicles that go stale immediately. We wanted to prove that a small team could build genuinely semantic gift retrieval over a live product catalogue, end to end, and run it in production.
02 Our Approach
We treated it as a full data engineering problem, not just a search box. A pipeline ingests Amazon product data, an LLM enriches each product with synthetic gift-style queries and a giftability score, and every one of those synthetic queries is embedded so retrieval can match on meaning rather than exact wording. Retrieval itself fuses multiple signals rather than trusting embeddings alone.
03 The Solution
A Django 5 site backed by Postgres with pgvector: an LLM enrichment pipeline generates per-product "facets" (synthetic gift queries, occasions, recipients, interests) with 512-dimension embeddings, run through a custom Postgres-based job queue so no separate broker is needed. A visitor's free-text query is matched using Reciprocal Rank Fusion across four independent arms: HNSW cosine similarity on the embeddings, Postgres full-text search, trigram similarity for typos, and controlled-vocabulary overlap. The result renders as curated topic landing pages with Amazon affiliate links, never showing prices to stay compliant with Associates rules.
04 The Result
The live site at regalosmejores.com takes a sentence like the one above and returns genuinely relevant, gift-specific picks instead of generic keyword matches. It's a working proof that embedding-based retrieval, paired with an LLM enrichment layer over a real product catalogue, meaningfully outperforms keyword search for high-intent, natural-language discovery, a pattern directly reusable for client search and recommendation work.
Tech stack
More case studies
Ready to put AI to work in your business?
Book a free 30-minute strategy call. We'll map the highest-ROI AI, data, and automation opportunities for your business, no obligation.
No obligation. 24h response. NDA on request.