Services
Independent, statistically sound search and RAG evaluation where a wrong result is expensive. Fixed-fee measurement when you have data.
Prices in EUR, excluding VAT. Prices in GBP and USD on request.
Asynchronous; one 30-minute walkthrough
An independent reading of what your search numbers mean for your decision, for teams with a plugin export, a query log that shows the results, or 30 to 100 questions written by their experts. From a log, a Readout needs no labelling; from expert questions, each question needs the document that answers it.
Deliverables
- An independent reading of what the numbers mean for your decision
- What to measure first; for document systems, which layer (extraction, retrieval or citation)
- A labelling plan sized for your case (sample size, who labels, scale)
- A fixed quote for the Evaluation
- The work behind them: a coverage and data-quality check, zero-result and failing queries grouped by cause, head versus tail, and query form
- A short written readout and CSVs of the top and failing queries
Asynchronous plus two calls
One decision measured, for example keyword versus semantic versus hybrid search, before versus after a parser, embedding or reranker change, or whether a RAG system's citations hold up.
Deliverables
- A stratified query sample
- Graded judgments by your domain experts using TensorOpt guidelines, with agreement measured on an overlap
- Candidate configurations run on staging or inside your environment
- nDCG@10, precision, recall and MRR (plus citation accuracy for RAG), with paired bootstrap confidence intervals, permutation tests and Benjamini-Hochberg over a declared comparison family
- Per-query wins and losses
- A report, the judgment set (you keep it), reproducible harness configuration and a walkthrough
For scanned or OCR-processed documents
Document systems where accuracy can be lost before retrieval starts.
Deliverables
- A labelled sample of pages or fields
- Character and field error rates with confidence intervals
- Where errors cluster (handwriting, tables, corrections)
Capped at 8 hours a month; paid monthly in advance
Teams that change their search or RAG system regularly and want each change measured against the same judgment set.
Deliverables
- Re-run the judgment set against the current configuration
- Evaluate up to 2 proposed changes a month
- One 30-minute call a month
- Quarterly: add about 50 new queries from a fresh log to track drift
Private, for one team, up to 12 people. Statistics-only version, 2 online half-days: EUR 4,000. Materials are built when the first buyer signs: 4 to 6 weeks' lead time
Teams that can build and change their system but lack statistical rigour.
Deliverables
- Query sampling and judgment sets
- Metrics and when each misleads (nDCG, MRR, recall@k, citation accuracy)
- Labeller agreement
- Calibrating an LLM judge against human labels
- Bootstrap confidence intervals, paired tests, multiple comparisons, sample size and power
- Reporting, and when first-party results need independent evidence
- Labs on the open-source harness, with public data or a sample of the team's own data
The booking buttons open our booking page on Calendly, a US service; it asks for your name and email. Prefer email? info@tensoropt.ai
Payment terms
Standard terms
- Readout: 100% on signature, by card (Stripe) or SEPA. Credited in full against an Evaluation booked within 60 days.
- Evaluation: 50% on signature, 50% on delivery, net 15.
- Retainer: monthly in advance.
Enterprise buyers (banks, insurers, public bodies and anyone with vendor onboarding)
- Invoice against a purchase order, net 30.
- Work starts when the purchase order arrives, not when payment does.
- The Readout credit applies the same way.
Vendor: Crafting Software Development and Consulting LLC, Templom u. 15., 2161 Csomád, Hungary. Company registration number 13-09-205369. EU VAT ID HU27924534. Reverse charge for EU and UK business clients; US invoices are outside EU VAT scope. Hungarian law governs the SOW unless negotiated.
Frequently Asked Questions
Have a question about your system?
Book a 30-minute call to discuss your current setup and which evaluation step makes sense.
Opens our booking page on Calendly, a US service; it asks for your name and email. Prefer email? info@tensoropt.ai