Built by physicians,
for the radiology routine
Laudos.AI is a Brazilian healthtech founded by physicians. We build the post-imaging layer of radiology: everything that happens between the acquired image and the signed report, dictation, structuring, review, classification support and critical-finding communication.
Our founding conviction is that AI in medicine must be assistive. The system proposes structure, terminology and a draft; the radiologist reviews, edits and signs. That design is not a limitation, it is the product: it is what makes the gain in speed compatible with clinical responsibility and with Brazilian regulation (CFM Resolution 2.454/2026).
We publish our evidence. RadCommons is our versioned corpus of classification systems; LaiBench is a public, blinded benchmark from finding to report. Median time from editor to signature is 52 seconds, measured in production over 5,200 reports in a 30-day window, with the methodology published.
The company is based in São Paulo, Brazil, with data residency in the country and a designated DPO. If you want to see the product on your own workflow, book a demonstration.
Four modules, one clinical flow
Voice dictation that becomes a structured report, in your template and your terminology.
Critical findings communicated with SLA, escalation and a closed loop until acknowledgement.
Classification suggestions (BI-RADS, LI-RADS, PI-RADS, TI-RADS) for the physician to review.
Structured reports per modality, with the institution’s template as a governed asset.
Plugs into your existing flow
HL7 v2, FHIR R4, DICOM-SR and DICOM Modality Worklist: Laudos.AI connects to your PACS, RIS and worklist without replacing anything. Deployment starts small and auditable, with a dedicated engineer and a 30-day pilot on your real routine.
Governance first
The AI proposes; the physician reviews, edits and signs. No autonomous reports (CFM 2.454/2026).
Sensitive health data stays in Brazil, under LGPD, with a per-exam audit trail.
RadCommons and LaiBench: our classification corpus and public benchmark, open to scrutiny.
Measured in production: n = 5,200 reports over a 30-day window.