AI for scientific research and labs
Ultramarine Work drafts Specific Aims and DMS plans, builds DOE run sheets and IQ OQ protocols, and edits manuscripts. It works on the files already in your lab folder.
R01-specific-aims.docxExampleSpecific Aims: Hypoxia and Persister Formation in Pseudomonas aeruginosa Biofilms
Aim 1
Define how hypoxia controls persister formation in mature biofilms. We will quantify persister fractions at four oxygen tensions in flow-cell biofilms and test whether deleting the anr regulator abolishes the hypoxic increase. Reviewers asked for a clinical anchor, so Aim 1 now includes 24 sputum isolates from the cohort.
Aim 2
Test whether restoring oxygen resensitizes biofilms to tobramycin in a murine chronic infection model. Our preliminary data show a 3.2-fold drop in surviving CFU when hyperbaric oxygen precedes tobramycin (Fig. 3). The power analysis in Approach 2.4 sets 12 mice per arm to detect a 0.5 log10 difference.
CheckedCritique 3 on the Aim 2 sample size is answered. Critique 5 on sex as a biological variable is still open.
How it works
- 1
One folder per project
The project folder holds the NOFO or journal guide, ELN and instrument exports, the manuscript with its figures, prior grants and protocols. Ultramarine Work goes through it and asks before reading a site on the web for the first time.
- 2
Ask in plain language
You ask for the Specific Aims redraft, the OQ protocol or the DOE run sheet. What comes back is a Word document, a workbook with live formulas or a slide deck, and anything it could not find or reconcile is flagged rather than guessed.
- 3
Verify, then decide
Every number and statement names the file it came from, so you can check citations against PubMed and tie results back to raw data. You, your QA unit or the PI then decide what goes to the funder, the journal or the record.
Ask it for
“Read the FOA and our preliminary data summary, then redraft the three Specific Aims of our R01 resubmission to answer the critiques in the summary statement.”
R01-specific-aims.docx“Build the modular budget justification for our 5-year R01, direct costs of $248,900 a year, fringe at 31.5% and indirects at 62% of MTDC.”
R01-budget-justification.xlsx“Draft my current and pending (other) support from my CV, the ORCID export and the grants list, with person-months and overlaps flagged.”
Davis-CPOS-draft.docx“Respond to the two reviewers’ comments point by point, and mark where each change lands in the text.”
Reviewer-responses.docx“Build a 2^(5-1) fractional factorial run sheet for the formulation screen: temperature, pH, stirring rate, protein load and hold time.”
Formulation-DOE-run-sheet.xlsx“Draft the OQ protocol for the new HPLC. Tie the acceptance criteria to each requirement in the URS, and add test steps and a deviation log.”
HPLC-OQ-protocol.docx
Inside the Scientific Research & Labs mode
Research grant proposal
Structures Specific Aims and research strategy against the NOFO’s page limits and review criteria, critiques your draft, and drafts the RPPR sections.
Grant budget and justification
Builds the multi-year budget justification with fringe, escalation and your negotiated indirect rate, and flags costs 2 CFR 200 will not pay for.
Biosketch and current pending
Prepares biosketch and other-support content from your CV and grants list, ready to carry into the SciENcv forms NIH and NSF now require.
Data management and sharing plan
Pulls together the data types, standards, repositories and access controls the NIH format page and NSF plan guidance ask for, plus the release timeline.
Scientific manuscript
Drafts and edits manuscripts, abstracts and posters, checks a paper against CONSORT or STROBE, and writes the cover letter.
Experimental protocol
Plans the experiment before it reaches the bench: hypotheses, design and controls, replicates, randomization, blinding, materials and procedure.
Literature and evidence review
Runs reproducible searches over the PDFs in your folder, then builds a review with a PRISMA flow, study appraisal and GRADE certainty ratings.
Statistical analysis plan
Sets estimands per ICH E9(R1), then the models, multiplicity, missing-data handling and power justification, so the analysis is pinned down before data come in.
Design of experiments
Turns a factor list into a fractional factorial or response-surface design, run sheet and analysis approach included.
Peer review report
Where the journal allows it, drafts the referee report against the journal’s stated criteria and keeps the confidential comments to the editor separate.
Invention disclosure
Assembles what the tech-transfer office expects, from the technical description and supporting data to prior art and public-disclosure dates.
Model evaluation and documentation
Curates a test set that mirrors deployment data, reports baselines, subgroup errors and calibration, and writes the model card a reviewer can question.
Verification test plans and reports
Builds DVP&Rs that tie every requirement to a test with acceptance criteria, then writes the report with uncertainty and deviations.
Technical requirements and specifications
Writes requirements you can verify, each with testable acceptance criteria, and traces them through specs to tests.
Validation protocol and report
Drafts IQ, OQ and PQ protocols for a new instrument under GAMP 5 and Part 11, and method validation per ICH Q2(R2).
Root cause analysis and CAPA
Structures OOS and deviation investigations that test each hypothesis against evidence and say how CAPA effectiveness will be checked.
Safety programs and hazard analysis
Writes the chemical hygiene plan 29 CFR 1910.1450 requires, plus site-specific safety plans and job hazard analyses for the riskier tasks in your lab.
Quality system audit and management review
Checks your lab’s practices against ISO 17025 or GLP clause by clause and drafts the management review inputs the standard expects.
Business case and capital allocation
Puts numbers on the new sequencer or the facility expansion, works through build versus buy, and states the trade-offs.
- Office skills
- Excel, Word, PowerPoint, PDF
- Tools
- Read, edit and write files, Search files, Run code, Search and read the web, View images
- Connectors
- SharePoint, Google Drive, Gmail, and additional custom connectors
- House style
- Your fonts, layouts and rules, on every file
Who it’s for
Life, physical, computational and ML scientists, postdocs and PIs in universities, institutes and industrial R&D, and the lab, core facility and QC lab managers who run their labs.
Questions
Can reviewers use AI for NIH grant peer review?
No, and NIH has said so plainly. NOT-OD-23-149 bars reviewers from uploading application content, critiques or original concepts to generative AI, and it calls that a breach of peer review confidentiality. ICMJE asks the same of journal reviewers. Manuscripts and proposals under confidential review go into an AI tool only where the journal or funder allows it, and that judgment belongs to you and your institution.
Will NIH reject an application it thinks AI wrote?
It can, under a policy in force since the September 25, 2025 receipt dates. NOT-OD-25-132 says applications or sections substantially developed by AI are not the applicant’s original ideas and will not be considered, and NIH is running AI detection to find AI-generated content. AI may still assist limited aspects, but plagiarism and fabricated citations are named risks. The hypotheses, preliminary data and logic stay yours, and every file it touched still needs your review.
Does it invent citations or make up data?
Yes, citations and numbers in a first draft can be invented, so treat every reference as unverified. The skills have it cite the source file behind each statement and note what it could not find or reconcile instead of filling the gaps. Fabricated references are the most damaging failure an AI draft can carry in this work, because reviewers search PubMed. Before anything leaves your folder, every citation has to be checked against the actual paper.
Can I paste a draft invention disclosure into it?
No, not until your tech-transfer office signs off, because that call is theirs. University IP offices warn that entering invention details into an AI tool may count as public dissemination and can destroy patentability in many countries, and only your institution can say where it draws that line, since whatever is in the folder is processed on Paper Instruments’ servers.
Where does our lab data go?
Ultramarine Work changes files only in the project’s folder. When a request needs it, it can read files elsewhere on your computer, never passwords, keys or app data, and you can turn that off in settings. The model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Whether patient or participant data may be processed at all is an IRB decision, and many institutions bar it even de-identified. Details are on our compliance page.