Rafiq Satria

I’m an informatics student and freelance AI + backend engineer who ships production code and starts from the problem to reach the right solution. I build systems that stay reliable when a model gets something wrong, pairing deterministic logic with graceful fallbacks instead of trusting AI output on faith.

In my recent work, I owned an end-to-end recommendation service—from data ingestion to a live API—architecting it as a deterministic retrieval core with a separate generative layer so the reliable path never depends on the model being right. I designed the LLM layer to “fail open,” added hybrid retrieval with Qdrant (dense embeddings + BM25) and rank fusion with cross-encoder reranking, and improved deployability and observability using CPU-only deployment patterns plus Prometheus, including fuzzy spell correction and automated indexing.