Search the literature from one question.
The research workspace searches PubMed and ClinicalTrials.gov together and lists every record it finds, each linked to its original page.
Available · research workspaceHOSPITAL WORK, CONNECTED THROUGH AN AGENT
Early access · research workspace liveThe research workspace returns the records it found, each linked to its PubMed or ClinicalTrials.gov page, so a researcher can read the originals. Synthesis across studies is a goal, not a current feature.
Literature search and investigator-edited protocols are available. Automated synthesis, ethics submissions and screening remain planned.
The research workspace searches PubMed and ClinicalTrials.gov together and lists every record it finds, each linked to its original page.
Available · research workspaceThe result is a source index and a cited evidence brief. Coverage is incomplete and abstracts can omit detail, so a researcher checks the original.
Available · source-linkedCompare findings, reconcile conflicting results and show how strong the evidence is. Planned; the workspace does not do this yet.
Goal · not available yetCreate an editable working study protocol from selected sources. Investigators write and review the study; automated drafting and ethics submissions remain future work.
Available · investigator draft editorCheck permitted, consented or de-identified records against a study’s criteria and hand the team a list to confirm. Planned.
Goal · not available yetA supervised agent system that logs what it read, what it changed and who approved it, so a study’s reasoning can be audited and reproduced. This is the design goal we are building toward.
Goal · supervised and auditableTranslational medicine is throttled less by a scarcity of ideas than by the friction that attends their examination: the interminable synthesis of heterogeneous literature, the laborious reconciliation of protocols with regulatory and ethical strictures, and the attritional curation of data whose provenance must remain beyond reproach. NutritiScan aspires to furnish medical schools with a supervised agentic apparatus that absorbs this friction, so that investigators may devote their finite attention to the questions that only human judgment can adjudicate. This is an aspiration and not an extant capability; its realization will be bounded by validation rather than by ambition.
Agents would survey and triangulate primary studies, trials and guidelines, reconcile discordant findings, and expose the strength and provenance of each inference, so that no conclusion rests upon an unexamined citation.
Aspiration · citation-groundedPre-specified analyses, versioned datasets and an immutable record of every transformation would render a study’s reasoning legible to a sceptical reviewer, and its results reproducible by an independent group.
Aspiration · auditable by designBy compressing the administrative interval between a promising finding and a properly governed protocol, we hope to hasten, and never to circumvent, the passage from laboratory observation to clinical inquiry.
Aspiration · oversight preservedOur goal is to help medical schools shorten the time between a clinical question and a carefully checked answer. Much of that time goes to reading, paperwork, screening and cleaning data, which are the parts an agent can prepare. It should never go into skipping ethics review, consent or oversight. Today NutritiScan offers only the research workspace; the steps below are what we are working toward.
Search, read and compare published studies, trials and guidelines. Summarize them with citations, and flag conflicting results and weak evidence so a researcher can check every source.
Help turn a clinical question into a clear hypothesis, outcomes, and inclusion and exclusion criteria, and point to similar earlier studies and the gaps they leave.
Prepare protocol drafts, consent-form wording and ethics-committee submissions from templates. Investigators edit them, and the committee decides.
Using only permitted, consented or de-identified data, check records against the criteria and give the study team a list to confirm. No participant is contacted without approval.
Check collected data for missing values, unit mismatches and protocol deviations, and keep a log of every change so the dataset can be audited.
Draft the statistical analysis plan, run the pre-specified analyses and prepare tables and manuscript drafts, with each number traced to its source. Statisticians and authors own the conclusions.
Research on people carries duties to the people in it. Agents would prepare and check work; researchers, ethics committees and participants keep the decisions. We will publish a speed claim only after it has been measured with a real research team.
Clear answers about what NutritiScan does today, what it is building toward, and where people stay in charge.
NutritiScan is building an AI assistant for medical schools and hospitals: a supervised agent system that helps investigators with literature review, study design, approvals paperwork, participant screening, data checks and reporting, and later helps hospital teams with records, documentation, handoffs and follow-up. It is in early access. Health chat, the research workspace and the Mac preview are available today.
The research workspace searches PubMed and ClinicalTrials.gov, returns cited evidence and supports an investigator-editable study protocol. The Mac preview adds PDF/PMC paper reading, exact passage links, CSV/FHIR inspection, reviewed descriptive analyses, reports and a terminal. Health chat remains available for general educational health questions. Automated protocol drafting and participant screening remain goals.
No. Researchers, statisticians and ethics committees keep every scientific and ethical decision. Ethics approval, informed consent, choices about participants and the interpretation of results stay with people.
NutritiScan does not run studies or offer a patient-record workflow. Desktop project files and saved history stay on your Mac; requested content goes to the model provider. Research searches go to PubMed and ClinicalTrials.gov. Use public or de-identified research data.
Not yet. Search coverage is incomplete, abstracts can omit important details and nothing here has been validated in a research setting. Check the linked source for any record you rely on, and we will publish speed or accuracy claims only after they have been measured with a real research team.
Health chat remains an important part of NutritiScan and is available from the homepage as an educational tool. It does not diagnose, prescribe or replace a clinician, and its messages are sent to our server and may be processed by a configured AI provider.
Research searches are sent to PubMed and ClinicalTrials.gov, so keep names and identifiable information out of them. Messages in the earlier health chat are sent to our server when you submit them and may be processed by a configured AI provider. NutritiScan does not use patient records for model training, and no study data or patient records are held today.
Read how your data is handledEarly access · research support only · people keep the decisions.