Private scope
Your owner and project boundary is established before research begins. No cross-owner or cross-project synthesis.
Whether you are learning in MBBS, training in MD, practising, or building research, start with the observation, uncertainty, or problem you cannot let go of. HiDoctorly turns it into a searchable question, an honest answer to “has this already been done?”, and — when appropriate — a protocol draft you can review with your team and ethics committee.
Nothing you write is readable by another clinician, pooled with another doctor’s work, or sent to any external model. That is enforced in code, not promised in a policy.
Every private question becomes an accountable research artifact: scoped to you, grounded in retrieved evidence, checked claim by claim, and recorded as it moves through the chain.
Your owner and project boundary is established before research begins. No cross-owner or cross-project synthesis.
Retrieval tier, source, date and licensed-content handling stay attached to the research run.
Claims bind to sources and pass the HI-EI-AI checks before an exportable record is issued.
Artifact hashes, approvals and audit events make the path reviewable after the answer is written.
You do not need to know how to phrase a MeSH query, and you do not need a research methodologist sitting next to you. You need forty minutes and one observation you cannot let go of.
Open a project, and type the observation in ordinary clinical language — “younger MI patients in the last two years seem to be presenting later than they used to.” No template, no keywords, no structured fields. Strip the identifiers before you paste; the system checks and stops you if you forget.
Run the cascade. You get papers with design and year, and a synthesis where every claim shows the source behind it. This is the step that either kills your idea in twenty minutes or tells you the gap is real — and either outcome is a good use of your afternoon.
Star the papers you already trust. They become Tier 1 — searched first, every time, and never sent anywhere. After two or three projects the cascade starts from your own reading list rather than from cold.
Ideas come scored for feasibility in your setting, not in an ideal one. A design needing 400 cases a year is marked as such if your unit sees 90. Pick against your real case volume and your real follow-up rates.
Export the protocol with the governance record attached — objectives, population, outcomes, sample-size assumptions, statistical plan, bias table, limitations. Edit it as a clinician. It is a strong first draft, not a submission.
Your corpus is yours alone. No cross-owner flow, no cross-project synthesis, nothing sent to an external model. The only thing that ever leaves for outer-tier search is a derived query string — never your case text.
What it will not do. It will not tell you what to prescribe, start or stop. It will not diagnose. It will not tell you your idea is the first in the world — it cannot know that, so it says “based on searches through this date, limited evidence was found” instead. If you want a tool that flatters your hypothesis, this is the wrong one.
Human-in-the-loop by construction. Nothing advances to the next stage without your signed approval, and every approval is written to an audit trail with the artifact hash it signed over.
Write what you saw, the way you would tell a colleague in the corridor — “the last dozen or so post-CABG diabetics on empagliflozin seemed to mobilise faster than I expected.” No structured form, no keywords. De-identify it: no names, no UHID/MRN, no dates of birth, no phone numbers. A PHI scanner runs before anything else and, if it finds one, intake stops and points at the exact text so you can strip it. HI-EI-AI then hands back your observation as a PICO — population, intervention/exposure, comparator, outcome — plus a three-sentence paraphrase in plain clinical English.
The cascade runs a query ladder built from your framing — tightest search first, broadening only until it has enough to stand on. You get a source list with study design, year and where it came from, and a synthesis in which every sentence is bound to the papers it rests on. Expect a genuine answer to “has this already been done?” If the honest answer is “partly”, it will say partly.
Candidate studies you could actually run, each with a proposed design — retrospective cohort, case-control, prospective, registry, diagnostic accuracy, prediction model — and each scored three ways: how new it is against what Tier 1–2 found, how feasible it is in your own unit given your case volume and what you can realistically measure, and how publishable it looks. The feasibility score is the one that saves you a wasted year.
A draft in the shape your ethics committee expects: objectives, population with inclusion and exclusion criteria, exposure, comparator, primary and secondary outcomes with how each is measured, sample size with every assumption written out where you can argue with it, statistical plan, bias risks with mitigations, ethics section, limitations. It is a first draft for a clinician to edit — not a submission-ready document, and it does not pretend to be.
Before anything leaves the system, a full validator chain sweeps the whole project — stance, evidence composition, cross-stage consistency, gate completeness, ownership and citation binding. This is the pass that catches the protocol that has quietly drifted away from the observation that started it, and the sentence with no paper behind it.
You will recognise the problem: a plain-English clinical question typed into a biomedical index returns almost nothing, because indexes match terms and not sentences. So retrieval here runs in four tiers, in order, working a query ladder from your framing — tightest first, broadening only until it has enough to stand on, then stopping.
The papers you have already read and kept, held inside your own project. Searched first, always, and never sent anywhere.
Peer-reviewed biomedical indexes, queried through a query ladder that walks from specific to broad until the corpus is grounded. A plain-English framing returns almost nothing from an index; the ladder exists because of it.
Retrieved only for papers you ask for by name. Bandwidth and attention are both finite; neither is spent without your instruction.
Society guidelines, trial registries and health-authority sources — reached through a private secured search engine that receives nothing but the derived search string. No case text, no patient-adjacent detail, no part of your corpus. Withheld entirely unless Tiers 1–3 produced grounding.
You are being asked to put your unpublished observation — the thing you have not told anyone yet — into a piece of software. So this is stated plainly rather than buried in a policy page.
Nobody reads your corpus. Not another clinician, not another department, not us. Every row is anchored to your owner id and every query is scoped to it. There is no admin view of your project text.
No external model ever sees it. Your observations, sources, ideas, protocols and claims are never sent to any third-party model. Reasoning runs against the endpoint your instance is configured with — in Private self, one inside your own estate.
No cross-project synthesis. Your projects are not mined against each other, and never against another doctor’s. Nothing you write becomes training data, an aggregate statistic, or a “similar research” suggestion to someone else.
PHI is refused, not stored. The de-identification gate blocks patient identifiers at the door, checked before the text is written anywhere. The safest posture is to hold no PHI at all, so that is the posture.
The outer ring gets a query string, nothing else. When Tier 4 runs, the Private Secured Search Engine receives the derived search terms and nothing more. No case description, no patient-adjacent artifact, no part of your corpus. That boundary is in code, not in a promise.
You can take it all with you. Every artifact is exportable with its governance record. There is no lock-in on your own research, and deletion means deletion of the artifact — the append-only audit ledger keeps only the fact that an action occurred.
Not a diagnostic aid. Not clinical decision support. Not a novelty oracle. Not a generic chatbot with a stethoscope. HiDoctorly supports the design of research, and the clinician remains the scientific decision-maker at every stage.
HiDoctorly is open to MBBS learners, MD students, practising clinicians, clinical researchers, and institutions where private validation space is critical. Tell us your role and setting, and we will route you to the right rollout.