LLM readiness check

For teams with an LLM prototype that already convinced people internally and now faces legal, privacy or a second tenant. Often a chatbot or an agent.

Wrong fit if you are still in the idea stage or have no operating model. Without a use case and without data, the check does nothing.

What you get

• Kickoff: which use case is in scope, which is not
• Review of the data and the current architecture
• Assessment of evaluation, observability, source grounding and cost
• Written report with an effort range
• Readout meeting

Five days

1. Day 1: Kickoff, use case, data sources, who decides go-live
2. Day 2: Retrieval and source grounding. What do the answers actually prove?
3. Day 3: Evaluation and observability. Can a bad answer be reconstructed?
4. Day 4: Operations, cost, tenants. What is missing for the second customer?
5. Day 5: Written report and readout

What I inspect

I check evaluation, source grounding, observability, retrieval, context, operations and tenants on your prototype. A claim without an openable source counts as a finding, not a detail.

Evaluation · Source grounding · Observability · Retrieval · Context · Operations · Tenants

Sample finding

Example from a healthcare RAG (anonymized): answers cited PDFs the user could not open. The pipeline fetched similar chunks, not the passage. Legal would have stopped it. In the check that would be written on day 2.

Afterwards

Kickoff, review, assessment, document, meeting. This is mostly for mid-size and regulated teams whose prototype has been stuck for months.

Afterwards you know whether and how the use case can go to production. The follow-on work is optional. No follow-up is booked into the price.