Using the Research Hub

Introduction

The Research Hub is Theator's AI-powered research workspace. It pairs you with a research assistant that brings four kinds of expertise to your study at once:

• a data scientist — builds your cohort, joins the data, and derives the measures your study needs
• a biostatistician — chooses the right test, runs it, and reports the result properly
• a clinician — interprets what the numbers mean and flags the subgroup worth a second look
• a medical writer — drafts in IMRAD structure, keeps the citations, and formats for submission

It works on your real data: the surgical intelligence in your Theator library — automatically captured video, structured surgical steps and events, and case metadata — alongside your own uploaded files and the published literature. Because intra-operative timings can be read from the annotated video rather than from the operative note, your measurements reflect what actually happened in the room.

What it removes: data, time, and finding the right people

Three things usually stand between a good clinical question and a finished paper, and the Research Hub is built to remove all three:

• Reaching the data. The answer is often already in your hospital, spread across the EHR, the operative note and hours of recorded surgery nobody has time to review. Theator gives you that surgery back as your own data, supercharged by AI: every recorded case is automatically annotated into structured surgical steps, events and intra-operative timings — measurements that exist in no EHR field and in no operative note, and that are unique to your hospital because they come from your operations. The research assistant uses that layer natively, inside the app, alongside the rest of your internal clinical data, the files you upload and the published literature — so what gets measured is no longer limited to what someone remembered to write down.
• The time each iteration costs. Traditionally, every turn of a study costs weeks: assemble the cohort by hand, chase the labs, wait on a statistician, then start over when a reviewer asks something new. Months → minutes — you ask in plain language, and the cohort, the joined data, the statistics and the draft come back while you are still thinking, so changing your mind costs a sentence instead of a month. When an iteration is cheap, the exploratory question and the subgroup check actually get run.
• Having all the right expertise at once. A serious study needs a data scientist, a biostatistician, a clinician and a medical writer. The assistant brings all four to the same conversation.

You stay the researcher. You set the clinical question, decide which patients count, and choose what belongs in the paper. The assistant supplies the reach, the rigor and the speed, and it guides you from exploration all the way to publication, one approved step at a time.

Finding the Research Hub

Open the Research Hub from the left sidebar, under Clinical Research. If you don't see it there, contact your Customer Success Manager to confirm your access.

The Research Hub home page is built around one message box: type what you want to work on and press send, and the Hub starts a new, untitled project for it (the project is named after your first exchange). If you would rather start from a prompt, pick one of the four starting points:

• Explore my data — start with a group of cases, look for patterns, then form a question
• Search the literature — find relevant papers and the gaps in what has been published
• I have a research question — start from a question, check the literature, then build a cohort
• Define a study group — filter surgical cases into the group of patients you want to study

A switch in the message box chooses which kind of project your message will start — Exploration or Research (see Two kinds of project below). Exploration is selected unless you change it; Explore my data and Search the literature select Exploration, and the other two starting points select Research. Beside the switch, a three-step strip — Explore → Focused analysis → Manuscript — shows which of those steps the chosen kind unlocks; hover a step for what it covers. To give a project a name and a description before you write anything, use the + button beside the switch ("Start a named project instead").

Your projects are listed in the panel on the left rather than on the home page.

New to the Hub? At the top right of the home page, New here? Take a quick tour starts a guided walkthrough of about a minute, and Example project beside it (after "or view the") opens the example directly. The tour also opens by itself the first time you visit the Research Hub in a browser. It runs inside the Example project described below, on a real finished study. Finishing the tour leaves you in the example to explore; leaving it early brings you back to where you started. You can restart the tour at any time with Take a tour in the ⋯ menu at the top right.

Start with the Example project

Under Shared with me in your project list, pinned first for everyone on the platform, is a project named "Kidney function after partial nephrectomy", marked with a Research label and an Example project tag — the Example project for short. It is a complete study, built inside the Research Hub exactly as you would build your own, and it is the fastest way to see what the Hub does before you start.

It follows one clinical question from beginning to end in a single conversation: does the time the renal artery is clamped during a partial nephrectomy affect how well the kidney works afterwards? You can read the conversation from the first message — a surgeon who isn't sure where to start — through choosing the question, locking the cohort, building the tables, running the analyses and producing the figures, to a manuscript drafted to a specific journal's structure. Every part of the Workbench is filled in along the way, so you can see what a finished Materials tab and Publication tab look like.

A few things to know about it:

• The data is synthetic. The patients, the case IDs, the measurements and the five "sites" are all invented and anonymized — nothing in it traces to a real person or a real hospital. The literature references are real published papers.
• It is read-only, and it is the same for everyone. You can open it, read the whole conversation and browse every tab, but you cannot chat with the assistant inside it, edit it, export it, copy it into a project of your own, or delete it. That is deliberate: it means nobody can break it and no two people ever see a different version of it.
• To do this with your data, start your own project. Type your own opening question on the Research Hub home page — the assistant will take you through the same steps on your cases.

Reading it end to end takes about fifteen minutes. If you only have a moment, read the first exchange and then jump to the Publication tab.

To find your way around it, collapse the project list on the left. The study's phases appear as numbered circles: select one to jump to where that phase starts, and hover the list to see their titles. Above them, Take the tour walks you through the Hub on this project, and the book icon shows or hides the short guide notes that introduce each phase in the conversation.

Don't see it? Look under Shared with me, below your own projects — open that section if it is collapsed. If the Example project still isn't there, contact your Customer Success Manager. Take a quick tour appears only where the Example project does, because the tour runs on it.

Two kinds of project: Exploration and Research

Every project is one of two kinds, and the project list keeps them in separate sections, Exploration and Research, with a third section, Shared with me, for projects colleagues have shared with you.

• Exploration — for exploring your data and Theator's with the assistant: advanced analysis, searching the literature, and collecting ideas worth pursuing. Its Materials tab works exactly as in a Research project, but its Publication tab is locked: there is no manuscript yet, so the tab shows the ways forward instead.
• Research — everything an Exploration does, plus analysis focused on your question and the tools that write the paper. Its Publication tab is open.

New projects are Explorations unless you choose Research when you create them.

Research Ideas

While you explore, the assistant can propose research directions, one approval card per direction, so you can keep one and leave the rest. The ones you approve are kept with the project under Research Ideas, on an Exploration's Publication tab. Select an idea to read it in full.

Starting a Research project from an Exploration

An Exploration's Publication tab shows a Start a Research Project card with two routes. Both keep everything already in your Workbench — tables, analyses, figures, files and references:

• Turn this project into research — the same project becomes a Research project, keeping its link, its conversations, its shares and its research ideas (listed afterwards under Additional Research Ideas). You confirm first, and this is one way only: a Research project cannot go back to being an Exploration. Only the project's owner sees this route.
• Branch off a new research project — the Exploration stays as it is and a new Research project is created beside it. Nothing is created until you approve what carries across. This is the better route when the exploration turned up several directions worth pursuing separately.

Each research idea offers the same two routes, so you can turn the project into research around that idea's question, or branch a new project off it. A Research project branched off an Exploration is listed under it in the project list, and the two project pages link to each other under Spun off from and Projects spun off from this.

Projects and conversations

Your work is organized into projects, listed down the left side of the screen. Use the search box to find one, or the + button to start a new one. The New project dialog asks for an optional name (it defaults to "Untitled project"), an optional description that is shared with the research assistant, and the project type, Exploration or Research.

A project is one study or research goal. Opening a project shows its front page: its name and a short description you can edit, its kind, how many conversations it holds, when it was created, who owns it and your access to it, followed by the conversations inside it and a message box to start a new one. Everything the assistant produces is filed underneath the project.

Each project holds as many conversations as you need — one for building the cohort, one for the literature, one for drafting the manuscript. All the conversations in a project share a single Workbench, so starting a fresh line of thinking never costs you the study you have already built up. Your cohort, tables and analyses stay live in the project rather than going stale in a spreadsheet.

Select New conversation inside a project to add another. The ⋯ menu next to a project offers New conversation, Rename, Duplicate project (a private copy of the project's Workbench and its latest conversation, as a new project of your own that isn't shared with anyone), Share and Delete; the menu next to a conversation offers Rename and Delete. Which options you see depends on your access to the project.

The workspace at a glance

Each conversation has two working areas:

• The AI research assistant (left) — a conversational assistant. Type any question into the message box at the bottom.
• The Workbench (right) — organized into two tabs: Materials, for everything the study is built from, and Publication, for the paper itself. It holds everything you have approved into the study, it is shared by every conversation in the project, and the manuscript is written from it.

Collapse the Workbench when you want the conversation wide; a folded tab on the right edge of the screen brings it back.

Everything you build is tied to the current project and saved as you approve it.

How you and the assistant work together

The assistant takes you from an open question to a finished paper: explore your data and the literature, focus the analysis on one question and cohort, then write the manuscript. The first step is open in every project; focused analysis and the manuscript come with a Research project.

A few things to know about how you collaborate:

• You approve every change. Whenever the assistant suggests content — a research question, a cohort, a table, an analysis, a research idea — an approval card appears in the conversation describing what's in it and why. Apply saves it to your Workbench; Discard keeps you working on it.
• Your own edits wait for Apply too. When you edit Workbench content yourself, such as the research question or the Paper Outline, a bar in the Workbench counts your unapplied changes, with Apply and Discard. If you try to leave with changes not yet applied, the Hub asks before discarding them.
• One step at a time — or Autopilot. By default the assistant pauses for your input between steps (Manual approval). The approval-mode control next to the message box switches a conversation to Autopilot: actions are approved automatically and the assistant works through the next steps on its own — toward your research question if the project has one, or to a complete article if it doesn't — until you turn it off. Autopilot applies only to the conversation you turned it on in, and it switches off when you reload the page. Creating a new project or turning an Exploration into a Research project always waits for your approval, even on Autopilot.
• The Workbench is the source of truth. Everything you approve lands in the Materials and Publication tabs, and the manuscript is drafted from there.

Working with the AI research assistant

Use the assistant the way you'd brief a research collaborator — discuss your goals and concerns as you would with a bright colleague. You can:

• Ask questions about your surgical data — for example, "What factors affect operation duration?"
• Ask it to build or refine a cohort based on inclusion and exclusion criteria you describe.
• Ask it to search the literature or summarize what it found.
• Ask it to draft your research question, methods section or abstract, then edit the result yourself.

Because the assistant reads the structured annotations on your recorded procedures, it can derive measurements directly from what happened in the operating room rather than from free text in the operative note.

The assistant works from the data in your Theator library and the context of your current project, so as you refine your question and cohort, its answers become more specific to your study.

The Materials tab

The Materials tab holds everything the study is built from, ready for the assistant to draw on when it drafts. What the finished paper actually uses follows the Sources you tick in the Publication tab — adding something here makes it available, it does not on its own put it in the article. The cards appear in this order:

• Cohort — the group of cases your study draws from: its name, the criteria that defined it, and a count of how many cases have video, annotations and operative reports. Upload a list of case IDs, or ask the assistant to build a cohort from filters such as procedure, venue, and dates. Your final analysis group is narrowed from here once you know which cases have everything the study needs. A cohort built from filters keeps growing as new cases arrive; Freeze pins it to the cases it selects today, and the card then reads "Frozen · pinned to these N cases on" the date you froze it. Open collection opens the cohort's cases in Theator, and Replace swaps in a different list.
• Analyses — statistical results such as comparisons, correlations and regressions. Each analysis records the method, the test choice, the result and a clinical interpretation, so you can cite it directly in the article.
• Tables — working data tables the assistant builds, such as a baseline "Table 1" describing who is in the study, and a results table holding what you measured. Open one to see its columns, units and how it was built.
• Figures — plots and diagrams, such as a cohort flow diagram or a chart of your results. Ask the assistant to produce one.
• Documents — separate deliverables the assistant drafts alongside the article, such as grant attachments (for example, Specific Aims). This card appears only in projects that can store them.
• Files — documents and data you upload, to share data or extra context for the research. You can upload PDF, Word, PowerPoint, HTML, TXT, Markdown, CSV, Excel (.xlsx), Parquet, and image files. Spreadsheet and data files (CSV, Excel (.xlsx), Parquet) can be loaded as working tables the assistant analyzes; other files are kept as reading material the assistant can draw on for context. Pin as governing document in a file's menu keeps that file in front of the assistant on every turn.
• References — literature the assistant finds (PubMed and others) plus any papers you save. Use + Add Reference to add your own.

Because Theator captures cases continuously and automatically, your cohort is drawn from a comprehensive, representative sample rather than a manually recorded subset — which helps reduce selection bias in your study.

The Publication tab

The Publication tab is where your study becomes a paper. In an Exploration it is locked and shows the ways forward instead (see Two kinds of project above). In a Research project it has two sub-tabs.

Preparation

The Preparation sub-tab sets out what the paper is, card by card:

• Research Question — the clinical question the project is built around. Type your own, or ask in chat and the assistant will help you shape one. Everything that follows is anchored to this question.
• Journal and Style — choose the Journal, the Article type and the Study design. The card names the reporting guideline that applies to your study design and, where the journal has one on file, its next submission deadline. To write without a target journal, choose Default format (IMRAD) in the Journal list: IMRAD structure, AMA Manual of Style. If a journal doesn't accept an AI-drafted manuscript of the article type you picked, the card says so and the assistant won't draft one — pick another article type, or keep it and write that part yourself.
• Paper Outline — the abstract-shaped summary the article is written from. Its fields, their order and its word limit follow the journal and article type you chose, and a word count in the card's header shows how close you are to the limit. Write it directly, or have the assistant draft it from the material you've built, and edit any field in place. Buttons on the card copy it to your clipboard, copy it as plain text for a journal's submission portal, or download the abstract alone as a Word file.
• Sources — controls what the article is written from. Everything in your workspace is listed here — the Paper Outline, research question, cohort, tables, analyses, figures, references and files — and you tick what belongs in the paper. The Paper Outline is required, so it is always included. Write article using sources then asks the assistant to draft from the items you ticked and to leave out the ones you didn't; it stays unavailable until the Paper Outline has content.
• Authors — the byline in order, with each author's degrees, affiliations, disclosures and roles (presenting, corresponding, industry employee). Edit an author, move them up or down, or remove them.
• Additional Research Ideas — research directions collected on the project, including any it kept from before it became a Research project. This card appears only when there are some, and nothing in it changes the project — each idea is a note for a study you could start separately.

Treat the source selection as direction to the assistant rather than a hard filter: it is instructed to use only your chosen sources, so review the finished draft and its citations before you rely on it.

Article

The Article sub-tab holds the drafted paper itself, written from the sources you chose, with a live word count. Edit it directly in the editor, or ask the assistant to revise any part of it.

Sharing, exporting and feedback

The ⋯ (Research Hub actions) menu in the top-right of the header acts on the whole project:

• Take a tour — replay the guided walkthrough on the Example project.
• Share Project — invite colleagues by email so they can open your research question, cohort, materials and draft. Choose Viewer (read only) or Editor (can add conversations and work in the Workbench). Everyone you've shared with is listed in the dialog, and you can change someone's access or remove them later. An editor who opens a conversation someone else started can't type into it, but can select Clone to my conversation to carry on in a copy of their own, in the same project.
• Downloads — files you have already generated, such as exported tables and mined data.
• Share Feedback — tell us what is and isn't working.
• Export — take your work out, in these groups:
  • Chat — Copy as Markdown or Copy as Plaintext copies the open conversation to your clipboard.
  • Project — Download project materials (.zip) downloads the project's materials in one file.
  • Publication — the article as a typeset PDF (Export Article as PDF (LaTeX)), a Confidential PDF (watermarked), a Word file, or LaTeX source (.tex), plus Export References as BibTeX (.bib) for your reference manager.
  • Blinded review — shown only when your chosen journal requires blinded review: the same PDF, Word and LaTeX exports with the author details left out.
  • Conference — an A0 poster (portrait or landscape PDF), Export Slides (PowerPoint), and the LaTeX source for the poster and the slides.

The article and conference exports are built from your Article sub-tab together with the byline and the references you have collected, so there is nothing to paste in. Your most recent edits are saved before the file is generated, and these options stay greyed out until the article has content. Export appears inside a conversation, and on a project's page when the project has materials to download; the Chat group appears only inside a conversation.

Before the article or the project materials leave the Hub, they are checked for patient and surgeon names, record numbers (MRN-style IDs), email addresses, full dates and counts below 11. An article or project download that still contains names, record numbers or email addresses is not downloaded; the message lists what was found, and you can ask the research assistant to remove or pseudonymize it, then download again. Full dates and small counts are flagged as warnings, because a data-use agreement may allow them.

These actions are unavailable in the Example project described above: it is read-only for every user, so sharing, exporting, copying and downloading are all turned off inside it. They work normally in your own projects.

Get help

Have a question about a study, or want hands-on support? Reach out to your Customer Success Manager or contact us at support@theator.io. Theator's clinical and support teams can help at every stage — from defining your question to preparing your manuscript for submission.

What's Next?

• Starting a Research Project from the AI Assistant — Ask a research question in the AI Assistant and continue it in the Research Hub
• Research with Theator: Overview — The kinds of clinical research Theator supports
• Data Export for Clinical Research — Export case data to Excel