Passion for Technology
GraphRAG has been in real enterprise use for a year. A balance of which question types it answers better than classic RAG, what it costs to operate, and when the extra complexity pays off.
Read more
Measuring RAG system quality is more subtle than it seems. Metrics, golden sets, and the most common evaluation mistakes.
Read more
Vector-only or keyword-only lose cases. Hybrid search combines BM25 + semantic via RRF. How and when.
Read more
After two years of RAG, clear patterns emerge: smart chunking, hybrid search, re-ranking, continuous evaluation. What to avoid.
Read more
The Assistants API offers persistent threads, tool calling, and file search managed by OpenAI. We examine when it pays off versus Chat Completions with your own logic.
Read more
Embeddings alone aren’t enough. A re-ranker over top-100 typically lifts precision 15-30%. When and how to integrate without drama.
Read more
Nomic released an open-source embeddings model rivaling OpenAI. When to use, comparison, and how to integrate in your RAG.
Read more
Gemini 1.5 Pro proved million-token context is real. What changes in RAG and architectures when the model can swallow an entire book.
Read more
OpenAI released text-embedding-3 with higher quality and the variable-dimensions trick. How to leverage what’s new without rebuilding your RAG stack.
Read more
pgvector 0.5 added HNSW and changed the conversation. When PostgreSQL with pgvector is enough, how to index well, and where it starts to hurt.
Read more