Phase 1
Intake
Upload or paste a manuscript. The product diagnoses what needs attention first.
From a rough draft to a submission-ready research paper.
The original prototype had the right modules, but the flow felt fragmented and low-trust. This redesign improves UX clarity by making status, academic risk, and next actions visible at every stage.
Primary improvement
Flow clarity
Each screen answers: where am I, what changed, what blocks submission?
Guardrail posture
No invented claims
Academic integrity becomes a visible product principle, not buried helper text.
Demo readiness
More believable
Less generic tab demo, more like a real desktop academic tool.
Active visual system
Warm editorial interface with research-grade restraint.
Claude-inspired parchment backgrounds, serif-led hierarchy, terracotta actions, warm ring shadows, and calm containers instead of generic SaaS cards.
Parchment canvas
Serif for authority
Terracotta CTAs
Visible trust signals
Make the workflow legible before making it beautiful.
The flow now reads like an academic process, not a collection of features. Each phase has a job, a user concern, and a product response.
Persona: non-native researcher
Desktop-first demo
Academic team project
Phase 2
Refine
Improve wording section by section while preserving the author’s claims and meaning.
Phase 3
Review
Catch contradictions, unsupported claims, and structure gaps before submission.
Phase 4
Cite
Resolve missing references, format style, and source completeness in one place.
Phase 5
Submit
Export final files, respond to reviewer comments, and move to a final-ready state.
User task
Bring in the draft
Upload DOCX, import sections, and run manuscript diagnosis.
Revise paragraph by paragraph
Apply tone, grammar, and clarity improvements with change tracking.
Check research logic
Scan for numerical conflict, terminology drift, and unsupported claims.
Fix references
Validate DOI, attach missing sources, and align to journal style.
Prepare final package
Export manuscript, reviewer reply, and submission checklist.
What they worry about
“Where do I start?”
The user needs prioritization, not a blank editor.
“Did AI change my meaning?”
Every suggestion must be inspectable and reversible.
“Is the argument consistent?”
Logic checks should feel academic, not generic proofreading.
“Is this citation format correct?”
Formatting and source completeness should be separated.
“Am I truly ready?”
The system should summarize what still blocks submission.
Product response
Readiness diagnosis
A first-pass scorecard surfaces top issues across structure, logic, and citation.
Transparent AI editing
Diff view, action history, and section-specific prompts reduce blind acceptance.
Academic guardrails
Claims without evidence and cross-section conflicts are elevated above style issues.
Citation resolution workflow
Missing DOI, style mismatch, and in-text usage all connect back to the paper.
Submission checklist
One clear end state replaces scattered “done” signals.
Never
No fabricated results
The system never invents data, claims, citations, or methodological details.
Always
Show the change
Every AI edit can be reviewed as a diff before acceptance.
Escalate
Flag academic risk
Contradictions, unsupported claims, and citation gaps are treated as primary issues.
Start with research context, not just an email and password.
The original onboarding was functionally correct but visually generic. This version frames signup as academic setup and makes the resulting personalization more credible.
Institutional email
Discipline setup
Preferred journal style
Sign in
Trusted research workspace
University account
Welcome back to your writing workspace.
Continue with your institutional email to access projects, citation memory, and reviewer response history.
or continue with Google
Onboarding flow
Clearer progression✓
Verify institution
Confirm organizational identity before access to collaboration and shared projects.
2
Set research profile
Choose discipline, native language, and common journal styles so AI suggestions fit the context.
3
Select plan & permissions
Differentiate individual researchers from team and institution admins.
4
Upload an optional writing sample
Use prior papers to tune tone without changing scientific claims.
5
Open your first project
Start from diagnosis, not from a blank empty dashboard.
Replace generic recent files with project readiness.
Instead of a typical file list, this dashboard tells an academic team what needs attention now: draft quality, missing citations, and which paper is closest to submission.
Prioritized work
Status over decoration
Academic context
Good morning, Tuấn
Your team has one paper close to submission and two that need attention.
The dashboard now leads with actionable readiness instead of abstract usage metrics. For a demo, this makes the product story much easier to explain in under a minute.
Completed papers
7
Archived and reusable as style references
In progress
2
Currently in edit and review stages
Style score
94%
Average writing quality across active work
Successful submissions
3
Tracked with reviewer-response history
Closest to ready
Federated learning in healthcare systems
68% submission-ready. Remaining blockers: 2 unsupported claims, 1 citation style mismatch, reviewer reply draft not started.
Nature format
2 claim flags
1 style issue
Deep learning for ECG classification
65% complete · logic conflict in sample count · abstract style issues remain
Next step: resolve review flags
Federated learning in healthcare systems
88% complete · citation polish and reviewer response pending
Next step: finish citation cleanup
Transformer models for NLP benchmark
Submission package complete · reviewer response template saved
Next step: monitor reviewer comments
Show what AI changed, why it changed, and what it refused to change.
The original editor was usable, but the hierarchy was flat. This redesign promotes section context, diff review, and academic guardrails above secondary controls.
Tracked changes
Academic-safe prompting
Section-based workflow
Abstract
Introduction
Methods
Results
Discussion
Conclusion
3 open issues
Abstract
Focus: language clarity, claim precision, and citation sufficiencyThis paper proposes a novel deep learning approach for electrocardiogram classification. The proposed method utilizes a hybrid CNN–LSTM architecture to extract both local morphological features and temporal dependencies.
Experimental results on the MIT-BIH dataset demonstrate that our model achieves 98.7% accuracy, outperforming existing state-of-the-art methods by a significant margin.
Tracked changes
presents → proposes
utilize → utilizes
Flag: “a significant margin” needs evidence or a more specific comparison.
Grammar
Correct subject–verb agreement
“The proposed method utilize” should be “utilizes” because the subject is singular.
Style
Use a stronger research verb
“Proposes” is more specific and academically conventional than “presents” for an original method paper.
Evidence gap
Claim needs support
“By a significant margin” reads stronger than the current evidence shown. Add a citation or replace with a measured comparison.
Quality snapshot
Current sectionB+Style
C+Logic
AStructure
BCitation
Academic integrity should feel like the product’s backbone.
This screen now reads as a review command center. The main change is hierarchy: hard risks first, structural opportunities second, and overall score pinned in a calm dark panel.
Hard risks first
Submission blockers
Review-friendly summaries
!
Numerical contradiction
Methods reports 100 samples, while Results refers to a dataset of 97 samples. This is treated as a submission blocker until resolved.
i
Terminology drift
The manuscript alternates between “deep learning,” “DL,” and “neural network” in ways that may confuse reviewers. Suggest choosing one primary term and one abbreviation pattern.
✓
IMRAD structure is intact
The paper follows the expected Introduction → Methods → Results → Discussion flow, so structural revision can focus on clarity rather than section order.
✓
No fabricated content detected
All 47 AI suggestions are classified as wording, sentence structure, or citation guidance. No new findings, metrics, or claims were introduced by the assistant.
i
Introduction lacks an explicit research gap
The transition from related work to your contribution remains weak. Add one concise paragraph clarifying what prior work misses and how this manuscript responds.
References should feel connected to the paper, not like a separate utility.
The redesign gives citation work a clearer narrative: add source, verify metadata, connect to in-text claims, then export in the journal style. This is much easier to demo than a plain list.
APA 7
IEEE
Vancouver
Chicago
Search DOI, author, title, or a missing in-text claim…
1
Verified
LeCun, Bengio, & Hinton (2015)
Deep learning. Nature, 521(7553), 436–444. DOI validated and already cited in the manuscript.
2
Verified
Vaswani et al. (2017)
Attention Is All You Need. NeurIPS 30. DOI present, metadata complete, and formatting matches APA 7.
3
Needs metadata
Zhang et al. (2023)
Cardiac arrhythmia detection using LSTM. Journal detected, but DOI missing. The source is also linked to one unsupported comparison claim in the Discussion section.
Institution admins need calm oversight, not a noisy control room.
The original admin view had the right ingredients. This redesign improves hierarchy by separating institutional health, user management, and quota risk into clearer visual layers.
Institution view
Quota awareness
User lifecycle
Total users
1,247
Up 12% this month
Papers processed
4,891
Usage is growing across teams
API calls / day
23.4k
78% of monthly quota used
Institutions active
18
3 enterprise renewals due this quarter
| User | Plan | Status | Recent activity | Action |
|---|---|---|---|---|
| Nguyễn Văn A nva@hust.edu.vn |
Pro | Active | Edited 3 papers this week | |
| Trần Thị B ttb@vnu.edu.vn |
Free | Active | Reached monthly manuscript limit | |
| Lê Minh C lmc@uit.edu.vn |
Admin | Locked | Permission issue under review |