I will implement rag vector search and ai semantic search for your ecommerce


About this gig
Stop losing sales to bad search.
If your ecommerce search returns no results when shoppers type natural queries instead of exact SKUs, you are leaving money on the floor. I implement production-grade RAG and semantic search that understands intent, not just keywords.
Real result: I am currently leading the AI search migration for one of Latin America's largest retailers (200+ stores, 1M+ daily users, 50K+ products), replacing Google Search API with a RAG-based system projected to save $500K per year.
What you get:
- Vector database setup (Pinecone, Weaviate, Qdrant, pgvector)
- - Embedding strategy and model selection
- - Hybrid search: keyword + semantic + reranking
- - Production deployment with monitoring and fallbacks
- - A/B testing setup to prove lift
Stack: Python (FastAPI), OpenAI / sentence-transformers, AWS, Docker, Kubernetes.
Why me: 10+ years building production backends at scale. Senior Platform Engineer with cross-team architecture ownership. I ship tested deliverables and document so your team owns the system after delivery.
Message me with your stack, catalog size, and what is broken about your current search. I reply within 1 hour with concrete next steps.
Get to know Martin Poli
Senior RAG and AI Search Engineer for Backend at Scale
- FromUruguay
- Member sinceMar 2020
Languages
English
My Portfolio
FAQ
Which vector database should I use?
Depends on scale, cost, and ops constraints. I help you choose between Pinecone (managed), Weaviate (self-hosted), Qdrant (open source), and pgvector (no new infra). The Architecture Review package includes this decision.
How much will the OpenAI embedding API cost?
For 50K products with OpenAI text-embedding-3-small, initial indexing costs roughly $1-2 USD. Query embedding is about $0.00002 per search. I include cost projections in Standard and Premium packages.
Can you integrate with my existing search backend?
Yes. Hybrid search combining your existing keyword backend with semantic vectors usually beats pure semantic. I integrate with Elasticsearch, Algolia, Typesense, OpenSearch, and Meilisearch.

