Clinical AI Evaluation by Dr. Rohit Sharma, MBBS

For teams shipping LLMs into care

Know where your health AI fails before a patient finds out.

Physician-written clinical test cases, an automated evaluation harness that runs in your CI, and a safety report your clinical lead can sign off. Built by a doctor who also writes the code.

Get a free 25-case Spot Audit See pricing

Spot Audit: 25 adversarial cases against your public product, findings memo in 48 hours, no cost.

What the cases are built to catch

Missed red flags

Chest pain, sepsis, stroke and paediatric warning signs buried in ordinary-sounding complaints.

Unsafe triage and dosing

Wrong urgency, weight-based dose errors, missed interactions and renal or pregnancy adjustments.

Confident hallucination

Invented guidelines, fake citations and diagnoses stated with more certainty than the history supports.

Regression after updates

The same case set re-run on every model or prompt change, so a fix in one place doesn't break another.

How it works

  1. 01

    Scope your risk surface

    A 30-minute call on your use case, users and failure modes. You get a written test plan the same day.

  2. 02

    Write and run the cases

    Clinical vignettes and adversarial prompts with a physician rubric, scored by multiple independent model judges and reviewed by hand where they disagree.

  3. 03

    Ship with evidence

    A safety-failure report ranked by severity, and the harness in your repo so every release is tested the same way.

Pricing

Spot Audit

Free

  • 25 adversarial clinical cases
  • Run against your public product
  • One-page findings memo in 48 hours

Eval Sprint

$3,500 fixed, 10 business days

  • 300-case eval set for your use case
  • Physician-written scoring rubric
  • Automated harness wired into your CI
  • Severity-ranked safety report
Book a sprint

Eval Retainer

$1,500 per month

  • Re-run on every model or prompt change
  • 50 new cases a month
  • Monthly safety report
  • 2-day turnaround on ad-hoc checks

Scope: evaluation and test design only. No clinical decisions are made for your users, and no patient data is handled without a signed data-processing agreement.

Who does the work

Dr. Rohit Sharma is an MBBS physician and the CTO of an enterprise AI platform running multi-agent pipelines for major retailers.

© 2026 Clinical AI Evaluation · hello@clinicalaiaudit.com