TensorOpt

Blog

Technical deep-dives on search, embeddings, RAG, and retrieval systems.

RAG12 min read
RAG That Works in Production

Most RAG hallucinations trace back to retrieval failures, not the LLM. Fix chunking, logging, and evaluation before upgrading models.

March 17, 2026

Evaluation10 min read
Measuring Search Quality: A Practical Guide to Reranking and Evaluation

Cross-encoder reranking delivers 25-48% retrieval quality gains. NDCG baselines turn search optimization from vibes into engineering.

March 10, 2026

Embeddings10 min read
Generic Embeddings Failed Us

Off-the-shelf embeddings get demos working. Domain-adapted models get production working. The gap is up to 9.3 nDCG@10 points.

March 3, 2026

Search10 min read
Hybrid Search: What Makes It Work?

BM25 isn't dead. Embeddings aren't magic. Running both and fusing intelligently is what production systems actually do.

February 24, 2026

Search10 min read
Why Your E-commerce Search Sucks

70% of top e-commerce sites can't handle synonym queries. The fix isn't a better algorithm. It's data you already own.

February 17, 2026