About TensorOpt
Independent, statistically sound search and RAG evaluation where a wrong result is expensive.
What TensorOpt does
TensorOpt is an independent search and RAG evaluation practice for organisations where a wrong result is expensive, with legal and financial work in focus. It measures search on portals and intranets (on Liferay DXP, starting with a free open-source plugin) and document search and RAG. Paid work is fixed-fee and time-boxed, and ends in a written report with classical information retrieval statistics: nDCG@10, bootstrap confidence intervals, and paired significance tests. The methodology is anchored in Designing Hybrid Search Systems (Leanpub).
Who runs it
Laszlo Csontos
Founder, TensorOpt
I am the author of Designing Hybrid Search Systems (Leanpub) and the forthcoming Measuring Search, and of the open-source Search Eval Logger for Liferay DXP and the open-source evaluation harness. Software and ML engineer, working independently for US and EU companies.
Background: five years at Hungary's largest retail bank; MSc in Computer Science, Budapest University of Technology and Economics (BME).
Why it exists
Search and RAG vendors publish their own evaluation results; that is structurally conflicted. TensorOpt is paid only by the client, is not affiliated with Liferay or any search vendor, and does not implement what it measures.
What stays internal
The engagement runbook (judgment-set construction templates and per-engagement configuration and scripts) is engagement IP and stays inside the practice. The evaluation harness is open source on GitHub. The published book covers the methodology in full. An Evaluation delivers the report, the judgment set, reproducible harness configuration and a walkthrough.
How to engage
Book a 30-minute call or review the services.