How Thesisroom works, in five sections.
This page is for the supervisor who wants to know whether our models can be trusted before their student turns to it. It is also for the candidate who wants to understand what the system is doing on their behalf.
1. The room.
Thesisroom is built for the defense, not for the writing. A tool that helps a candidate compose paragraphs ends up writing for them. A tool that pressure-tests their work after it is written holds them to a higher standard than they could hold themselves to alone. The room is the time between the draft and the viva, and that is where Thesisroom does its work.
Our models never produce thesis prose. They produce questions, marginal comments, and verification reports. The candidate writes the thesis. Thesisroom interrogates it.
2. The split.
Two kinds of work happen inside our models, and they are handled by two different systems. Facts about citations and the academic record are checked deterministically against five registries, with the registry response shown beside every flag. There is no language model deciding whether a citation is real.
Arguments, questions, and structural critique are handled by an AI model that operates only over the candidate's own thesis and its bibliography. It cannot invent sources. Between the two layers sits an audit step that confirms every AI claim points to a page in the thesis. Nothing reaches the candidate without a source.
The split is the whole product. Deterministic where determinism is possible. AI only where reasoning is required. The audit between them is what makes the AI output safe to read.
3. The sources.
Every citation in your bibliography is checked against five academic registries. Crossref returns the canonical DOI metadata. OpenAlex returns the relational graph: authors, institutions, related works. Semantic Scholar returns abstracts and the work's place in the citation network.
Retraction Watch returns the retraction record for the citation if one exists. DOAJ confirms whether the journal is a recognised open-access publisher. A citation that fails any of these checks is flagged with the exact registry response that produced the flag.
4. The provenance.
Every question Thesisroom writes carries a page reference back to the passage in your thesis that produced it. Every margin comment is anchored to a sentence or paragraph. Every citation flag shows the registry response side by side with what you wrote.
The candidate can verify everything our models say. So can the supervisor. Our models are not asking to be trusted on their word. They are asking to be checked.
5. What we refuse to do.
Thesisroom will not rewrite your text to evade plagiarism detection. The model has no access to a paraphrase tool, and the audit layer would block the output if it did. If your writing carries signals a detector might catch, our models flag them and ask you to revise. They do not revise for you.
Thesisroom will not fabricate citations, results, or participant data. The model is constrained to reason only over the bibliography you uploaded. It cannot recommend a citation that you did not put there.
Thesisroom will not produce thesis paragraphs. The structural critic returns marginal comments and revision prompts, never new prose. The writing stays yours, which means the defense stays yours.