Physician · Healthcare AI evaluation

Clinical judgment for trustworthy health information.

I evaluate medical claims through the lens of clinical practice, evidence quality and patient safety. My current independent project develops a structured way to audit plain-language summaries generated by AI.

01 / Selected work

Evidence, examined closely.

I work across clinical fact-checking, plain-language communication and error classification. The generator and auditor run end to end on development cases; the final evaluation panel has not been run.

Independent project · evaluation in development

Auditing AI-generated trial summaries

A lay summary can sound clear while overstating a benefit, changing a number or omitting a clinically important caveat. I built a generator and a separate auditor to compare summaries with a structured evidence table and classify those errors.

Two published trials, PARAGON-HF and SELECT, have field-traceable evidence tables and annotated base summaries. In PARAGON-HF, the primary outcome did not meet conventional statistical significance; exploratory findings must not be presented as confirmed efficacy.

The project defines 17 error codes (six commission, eleven omission) and has a four-case development panel: three cases with injected errors and one reviewed negative. The final panel has not been run, the reference standard has not been annotated, and no performance metric has been computed.

Evaluation workflow

  1. Extract published study findings into a field-traceable evidence table.
  2. Generate a lay summary from the table and inject a known error into selected development examples.
  3. Give a separate auditor the summary and the same table, without the source document or the inserted-error label.
  4. Compare coded findings with prespecified targets; a human-annotated reference standard is still needed for the final evaluation.

Method snapshot / PARAGON-HF

Explore a clinical claim

Development stage
17defined error codes
4development cases
Pendingreference standard & final metrics

Choose a case

A benefit claim that goes too far

Illustrative summary sentence

“Sacubitril–valsartan significantly reduced cardiovascular death and heart-failure hospitalizations.”

Reference evidence

The trial’s primary composite outcome had rate ratio 0.87 (95% CI 0.75–1.01; p=0.06). The prespecified primary test did not meet conventional significance.

Review finding

C1 · Unsupported benefit. This sentence claims a confirmed reduction that the primary result does not establish.

Why this matters

A reader could mistake an uncertain primary result for proven clinical benefit.

Constructed teaching examples, not recorded model outputs or measured audit results. Trial source: Solomon et al., NEJM 2019 ↗

02 / Communication

Writing that holds up to scrutiny.

Evidence translation · new

Obesity Explained

A public-facing evidence brief on obesity and its treatments. It separates weight change from clinical outcomes, pairs benefits with limitations and safety, and makes clear where trial results do — and do not — apply.

Read the evidence brief →
14auditable claims mapped to source evidence
13primary or regulatory sources
EUregulatory context distinguished from access
Plainlanguage without promotional framing
Plain-language summary

EMERGENT-3 / KarXT

A patient-facing summary of a schizophrenia trial, developed as an independent writing sample with attention to outcome hierarchy, adverse events and neutral wording.

Request the sample ↗
Evidence appraisal

Claims and uncertainty

My work checks whether a number, clinical conclusion or expression of certainty is actually supported by the underlying evidence.

Read my LinkedIn posts ↗
Medical publication

Case report

Co-author: Trotta PB, Rosa KG, Rosa LMA. “Myxoinflammatory Fibroblastic Sarcoma: Case Report.” Health Sciences Journal. 2015;5(2):139–147. The journal lists my surname as Rosa in its citation.

Read the publication ↗

03 / Background

Built in clinical practice.

Clinical medicine

Primary care, occupational health, medical examination and emergency medicine in Brazil. I currently practise in the Passa Quatro primary care system.

Education and languages

Medical degree, Faculdade de Medicina de Itajubá. Postgraduate certificate in Artificial Intelligence and Data Science in Healthcare at Hospital Sírio-Libanês, in progress (expected 2027). Brazilian Portuguese native; English professional working proficiency.