We connect to your existing search index using read-only credentials you provide. We run one baseline experiment — 100 auto-generated questions scored on answer correctness, context recall, and retrieval quality. No configuration comparison, no changes to your environment. You get a measurable baseline and a clear picture of where your setup currently fails.
We connect to your existing Azure AI Search index using read-only credentials you provide. We run one baseline experiment — 100 auto-generated questions scored on answer correctness, context recall, and retrieval quality. No configuration comparison, no changes to your environment. You get a measurable baseline and a clear picture of where your setup currently fails.
Delivered within 5 business days of credential receipt.
We never ask for credentials on this page. Here's the exact process after you submit.
Within 1–2 business days you receive a step-by-step email guide: how to create a read-only service principal scoped to your search service.
If your policy requires it, we sign a mutual NDA before any credentials are exchanged — entirely by email, typically under 24 hours.
Via encrypted channel. Used solely to run retrieval queries — no document download, no schema changes, no write access.
The experiment runs automatically. Once you receive the report, you delete the service principal. We confirm removal from our end.
We'll reach out within 1–2 business days.
A structured report — not a sales pitch.
One controlled experiment on your existing production setup — 100 auto-generated questions run against your current configuration. No changes, no tuning yet: this is your starting point.
Each question scored on answer correctness and context recall. You see how often the right content is retrieved and how often the generated answer is grounded in it.
What share of questions result in a severe failure — wrong or hallucinated answer. Failures are broken down by category so you know whether the problem is in retrieval, prompting, grounding, or context assembly.
The full results in a structured format — ready to compare against future experiments. This is your objective starting point, useful whether you go further with us or optimize in-house.
Not included in the free audit
These are part of the paid optimization engagement.
From request to report in 4 steps — typically completed within 5 business days of credential receipt.
We send the Access Guide by email. NDA by email if needed. You set up read-only credentials.
Day 1–2
We generate 100 test questions from your index content using AI sampling across topic clusters.
Day 2–3
1 baseline experiment run. All 100 questions scored on answer correctness, context recall, retrieval quality, and severe fail rate.
Day 3–4
Baseline report delivered by email. You revoke credentials — we confirm deletion from our end.
Day 5
What we access, what we don't, and what happens to your data.
PII note: The index should not contain unmasked personal data. If your index includes user-generated content with PII, we recommend using an isolated test copy with sensitive fields masked or excluded before the audit.
Everything you'd want to know before sharing access.
No. We can work with a read-only isolated test index — a copy of your production data without live traffic or user data. Mention your preferred setup in the form and we'll tailor the access guide accordingly.
We need a read-only API key or service principal scoped to your Azure AI Search service — specifically the Search Index Data Reader role. We'll send you a step-by-step setup guide by email after reviewing your request.
Yes, this is the preferred setup. An isolated test index with representative data gives you the same quality results without any production risk. Even a 20–30% content sample is sufficient for meaningful experiment results.
No. We issue retrieval queries and read the returned text excerpts — the same way your RAG application does. We don't download, copy, or store the source documents. Query logs are purged within 7 days of audit completion.
We'll sign a mutual NDA before any credentials are exchanged. Just mention it after submitting this form and we'll include it in the onboarding package. This typically adds less than 24 hours to the timeline.
You delete the service principal or rotate the API key in your Azure portal — takes 30 seconds. We'll send a written confirmation that we've removed the credentials from our systems. No further action required on your end.
The retrieval audit focuses on the search layer — Azure AI Search is the primary target. If you're using Azure OpenAI with On Your Data (which calls Azure AI Search underneath), the audit covers that setup as well. Let us know your stack when you submit.
That's a valid outcome. You'll receive the full results table showing your current configuration scored well across test questions — and you'll know exactly which configurations were tested. A clean result is still a useful benchmark.
It's genuinely free. We run the audit because it's the most honest way to show what we do. Some clients commission a paid deep-dive after seeing the results — but there's no obligation and no sales pressure.
Amazon Bedrock Knowledge Bases and Vertex AI Search are on the roadmap (Q3–Q4 2026). Submit the form and select your platform — we'll be in touch when your platform is ready.
Fill out the form above — it takes under 3 minutes. No credentials required today.
Request Free Audit