A writing instrument that refuses to invent evidence.
InkPhD is evidence-grounded infrastructure for medical dissertations and scientific papers. It drafts, cites, checks statistics and rehearses the defence — but it will not write a sentence your own sources cannot support. Everything it produces can be traced back, line by line, to a document you put there yourself.
The purpose: make a defensible manuscript the cheapest path, not the hardest one.
A doctoral manuscript fails for boring reasons. A citation that does not say what it was cited for. A survival analysis without a proportional-hazards check. A Methods section missing the ethics approval number. A results table that stops matching the text after the third revision. None of these are failures of intelligence — they are failures of bookkeeping, spread over three years.
General-purpose AI writing tools make that worse. They are fast, fluent and perfectly willing to produce a reference that does not exist. The purpose of InkPhD is the opposite: to make the rigorous route the fast route, so that the easiest thing to do is also the thing that survives peer review.
Retrieval first, generation second — and provenance all the way through.
Retrieval before generation
Nothing is written from the model's memory. Every paragraph is assembled from passages retrieved out of the library you built — your PDFs, your PubMed imports, your Zotero collection — and the passages travel with the text as provenance.
Numbered citations that resolve
Citations are emitted as strict Vancouver numbers and each one is resolved against Crossref or PubMed. A reference that cannot be found is flagged, never quietly formatted into something that looks real.
Claims are scored, not trusted
Each sentence is re-checked against the evidence that supposedly supports it. Unsupported statements are underlined in the editor while you write, the same way a spell-checker underlines a typo.
Human approval is the gate
No section is ever finalised automatically. You approve, the approval is recorded, and the export carries an explicit AI-disclosure line. That is what makes the output defensible in front of a committee.
Why it was built at all.
InkPhD began inside a clinical practice, not a software company. It started as private tooling for one medical dissertation: scripts to keep a reference list honest, to recompute survival curves after every data cut, to check a chapter against the reporting guideline it was supposed to follow. The writing was never the bottleneck. The bookkeeping around the writing was.
The same complaints came back from every colleague: the literature is too large to hold in one head, statistical software lives far from the manuscript, and the language editing services that are affordable do not understand methodology. Meanwhile paying someone to write the thesis is both common and corrosive — it produces doctors who cannot defend their own work.
So the goal was set deliberately narrow: keep the author in charge, remove the bookkeeping, and refuse to generate anything that cannot be traced to a source. The platform asks more of you than a generic AI writer does — your data, your sources, your judgement on every claim. That is the point. The result has to be genuinely yours.
What the platform actually does.
Document-centred workspace
Three panes — section hierarchy, scientific editor, intelligence panel — for ten document types: doctoral dissertation, research article, systematic review, meta-analysis, clinical protocol, grant proposal, case report, conference abstract, monograph and editorial.
Evidence library and discovery
Upload PDFs or import from PubMed, OpenAlex, Semantic Scholar, Crossref and arXiv; sync a Zotero collection; fetch open-access full text automatically by DOI through Unpaywall and Europe PMC.
Systematic review pipeline
PRISMA 2020 checklist, deduplication, AI pre-screening and an ASReview-style active-learning queue that reorders the remaining records after every decision, with a stopping rule and a recall estimate you can report in Methods.
Statistics on your own data
Upload a dataset and generate Kaplan–Meier curves with risk tables, Cox models with a proportional-hazards check, ROC/AUC, regressions and descriptive tables — each with the methods sentence and the figure legend written for you.
Reporting-guideline compliance
Eleven EQUATOR guidelines and 426 individual items (CONSORT, STROBE, PRISMA, CARE, SPIRIT, STARD and more), audited item by item with suggested fixes.
Consensus and citation mapping
A design-weighted consensus meter shows how many studies support or refute a stated claim; the citation network map reveals the classics your own references keep citing.
Defence preparation
Five reviewer personas interrogate the manuscript — methodologist, statistician, clinician, editor and sceptic — and a readiness score tells you what a committee will attack first.
Draft intake and writing plan
Upload a manuscript you already started; the platform splits it into sections, imports its reference list, audits the gaps and turns every gap into a tracked writing task that is AI-verified when you tick it off.
Publication-ready output
DOCX with native Word fields for figure and table numbering and cross-references, plus PDF, LaTeX, Markdown and HTML, and thousands of CSL citation styles.
Characteristics of the system.
Architecture. A React front end, a Python/FastAPI back end, MongoDB for documents and a per-project vector index for retrieval. Each project gets its own isolated index; no query ever reaches another author's material.
Retrieval. Hybrid search — dense embeddings plus lexical matching — over chunked full text, with the source, page and passage kept next to every retrieved snippet so provenance survives all the way into the export.
Models. Third-party general-purpose models are used through an API for generation, embeddings and transcription. We train nothing on your content, and your manuscripts are not submitted for model training.
Statistics. Real statistical libraries, not a language model guessing numbers: survival analysis, regression, diagnostics and plots are computed from your uploaded dataset and reported with the assumptions tested.
Verification. Identifiers are resolved against Crossref and PubMed; open-access full text is fetched where it exists; claims are re-scored against the retrieved evidence; reporting guidelines are audited item by item.
Safeguards. Research Use Only. Not a medical device, not clinical decision support, no diagnosis and no treatment recommendations. AI involvement is always explicit, every generated section is labelled, and the author's approval is recorded as metadata.
Data protection. GDPR by design: self-service export and permanent erasure, a published subprocessor list, an Article 28 DPA for institutions, strictly necessary storage only — no advertising, no analytics, no cross-site tracking.
Built for the people who have to defend the text.
PhD candidates in medicine
The people this was built for: one big manuscript, years of work, a committee waiting at the end.
Clinical researchers and residents
Case reports, cohort analyses and conference abstracts written between shifts, with the statistics done properly.
Systematic reviewers
Teams that need PRISMA discipline, screening that scales and an audit trail they can publish.
Supervisors and research groups
Shared standards across a department — the same guideline checks, the same citation verification, the same readiness score for everyone.
The interface is in English; generated content can be produced in twelve languages, including Bulgarian, so you can write for a local committee and publish internationally from the same project.
Who stands behind it.
InkPhD is owned and operated by International Sci Ink Press Ltd — an independent academic publisher founded in Sofia, Bulgaria in 2018 by Dr Ivan Inkov, a thoracic surgeon who still runs its editorial desk.
The house publishes two peer-reviewed, open-access titles: the International Journal of Medical Reviews and Case Reports and the International Journal of Surgery and Medicine. Between them they have registered roughly nine thousand Crossref DOIs, are curated by an editorial network of more than 180 members across 40 countries, and are archived for the long term in Portico and CLOCKSS. Every submission is screened for plagiarism, held to COPE, ICMJE and Declaration of Helsinki standards, and required to follow the EQUATOR reporting guideline that matches its study design.
That is the origin of this platform. Eight years of desk decisions produce a very clear list of why good research gets rejected: a citation that does not support the sentence, a survival analysis without its assumptions tested, a Methods section missing its ethics statement, a reporting checklist quietly ignored. InkPhD is that list turned into software — the editorial standard applied while the manuscript is being written, instead of months later in a reviewer's report.
The publisher also maintains other scientific-infrastructure projects; InkPhD is the writing instrument in that family. The platform, its source code and the InkPhD brand are the property of International Sci Ink Press Ltd. Your manuscripts, datasets and uploaded literature remain entirely yours — we claim no rights over what you create and we do not use it to train models.
Registered in Sofia, Bulgaria (EU) · company number 205414288. Full commercial and tax details are on the billing page, contact addresses in the privacy policy.