Built RAG-based personalized feeds, similar-event search, and grounded recommendations with embeddings and pgvector.
AI Recommendations / RAGCase study · AI & Backend Engineer
ActivityFinder
An AI-powered event recommendation and ingestion platform using embeddings, vector retrieval, semantic search, and automated data workflows.
KNN
personalized vector ranking
Team experience, including partner-led work. The role described here is ai & backend engineer; it identifies the implementation contribution rather than implying sole ownership of the product. For a new engagement, we agree the scope, responsibilities and release criteria before development begins.
01 · The challenge
The product problem
Turn heterogeneous event data into clean, searchable inventory and generate grounded, personalized recommendations without losing structured filtering or data freshness.
02 · Engineering
Architecture and implementation
Designed user-preference vectors and KNN-style ranking alongside structured filters.
Implemented re-embedding expiry and cleanup to maintain vector relevance.
Created n8n scraping and ETL workflows to collect, normalize, deduplicate, validate, and inject event data through the API.
03 · Stack
Technology used
FastAPIPostgreSQLpgvectorOpenAI
04 · Outcomes
Personalized vector ranking
Grounded event recommendations
Automated ingestion and validation
Planning a similar product?
Start with your users and one core journey. We can help shape the first release, choose a mobile or web approach, and connect design, engineering and launch support around the same outcome.
Next project
Confidential Fintech Platform
Fintech / Full-Stack