About TensorOpt
Intelligent search and retrieval, done right.
Our Approach
Search is a solved problem—until it isn't.
BM25 works great until users search for "running shoes" and expect trail runners, road shoes, and racing flats. Generic embeddings work great until your product catalog has terminology no model has ever seen.
TensorOpt exists because the gap between "search works" and "search works well" is where revenue lives. I combine traditional information retrieval with modern ML—hybrid architectures, custom embeddings, cross-encoder reranking—to build systems that actually understand what users want.
I measure everything. If I can't quantify improvement, I haven't improved anything. Every engagement starts with baselines and ends with measured lift.
Who's Behind This
Laszlo Csontos
Founder, TensorOpt
I'm a senior ML engineer specializing in search and retrieval systems. My day job involves building semantic matching systems, custom embeddings, and hybrid search architectures for production use at scale.
My focus areas:
- Fine-tuned embedding models
- Hybrid search (BM25 + semantic fusion)
- Cross-encoder reranking
- GraphRAG and knowledge graph retrieval
- Relevance evaluation and measurement
Want to work together?
Let's discuss your search and retrieval challenges.