Clinical trial intelligence
The patients your filters miss, with the evidence to confirm them.
Built for the trial, tuned for the field · No EHR integration
Deployment
Conventional platforms restart the rollout at every hospital. Krv builds the search once, each site runs it locally, and new sites come online in parallel.
Sites run our tools. PHI stays out of our AI, compute and servers.
The rollout · per hospital
Per-site governance loop — the queue before the queue.
Legal review for every data transfer, at every hospital.
A new vendor inside the perimeter means a new audit.
Custom interfaces, HL7/FHIR projects, IT tickets.
The only gate that remains — a copy, not a build.
One package, built once.
From learned pattern to a named patient
Everything left of the boundary happens on open-source data. Everything right of it happens inside the site, on records that never leave.
Standard filters look for what clinicians already know to search for. We start on broad, anonymous data to reveal the high-value patient cohorts everyone else misses—surfacing hidden patterns across care, documentation, and biology simultaneously. Built-in clinician review then validates these non-obvious structure into actionable clinical tools.
Learned cohorts are scored against the trial's criteria, with a rationale for each. Similarity guides who to look at. It never acts as an automated gatekeeper.
One lab value can't show whether a patient is improving, destabilizing, or drifting. Standard queries treat identical snapshots as the same patient. Our time-series models read the full trajectory to separate true candidates from likely screen failures.
We fine-tune custom models for the protocol on open-source clinical data—capturing complex signals across notes, labs, and trajectories—and package them to run offline at the site.
Drop a broad EHR export into our desktop app on the hospital's own machine. Fine-tuned model packs extract signals from notes and rank candidates 100% locally—fully private, with zero data transfer.
Investigators get a prioritized shortlist mapped to protocol criteria, complete with forecasted eligibility and clear explanations for every match. They confirm candidates instead of hunting for them.
Screening runs locally in a private desktop app on an approved export under site-specific data controls—no EHR integration and no data transfer.
The handoff to the investigator
Today · one site, adult SMA
A broad diagnostic-code query returns 60 charts. Two neurologists read every one of them.
Holly et al., JMIR Med Inform 2026 · Penberthy et al., J Oncol Pract 2012
With Krv · the same records
Ranked by the evidence behind each one, so review starts where it pays — and every position traces to a rule you can read.
PT-0417, review priority 1: A1c 8.2 percent, rising across the last four readings and estimated to enter the screening window within six months; one 41-day treatment gap, estimated around; eGFR 62, inside the protocol range.
Solid marks are measured history; faded marks are estimates, always shown as estimates. The output is prioritization, not eligibility.
Bring the protocol
We'll review the target population, searchable evidence, local requirements, and the most practical delivery path.
Review a protocol with usBLIND TEST · 9,172 RECORDS
/10092 out of 100
Case study: in a separate diagnosis-blinded evaluation, Krv identified 91.6% of known cases vs. 56.5% with the current industry-standard approach.