Sevivra

A collaborative scientific research platform and publisher: researchers and AI co-develop studies, accelerating discovery, improving rigor, and publishing transparent, reproducible research.

Comment

Inspiration

As a graduate student actively involved in academic research, I have experienced firsthand how inefficient and fragmented the publishing process can be. Critical parts of the system—such as peer review—depend on unpaid labor, while editorial decisions like desk rejections are often opaque and provide little constructive feedback, despite the significant value academic publishing generates.

At the same time, generative AI is increasingly used in research workflows, yet much of the existing publishing ecosystem remains resistant to work that transparently incorporates AI into its methodology. Sevivra was created to address these gaps by building an AI-enhanced academic publishing platform that streamlines collaboration, brings transparency to review processes, fairly compensates peer reviewers through a credit-based model, and actively supports the publication of AI-enabled research.

What it does

  • AI-assisted manuscript editor with a rich-text workspace, outline, review/diff flow, and auto-save.
  • Chat-based AI copilot that can respond and optionally return updated manuscript content.
  • Projects and manuscripts: create projects, manage project collaborators, track updates.
  • Comments and collaboration: inline comments tied to manuscripts, collaborator selection per project.
  • Publishing flow: publish a manuscript to a Sevivra journal specializing in AI-research, with AI “desk review” analysis.
  • Auth + admin approval: JWT cookie auth, user registration pending admin approval, super-admin tooling.

How we built it

  • Frontend: Next.js App Router + React 19, Tailwind + Radix UI, TipTap editor.
  • Backend: Next.js API routes with Zod validation, MongoDB via Mongoose.
  • AI integration: Gemini API via server-side curl and robust JSON/HTML parsing.
  • Auth: JWT cookies, middleware for protected routes, server-side user lookup.
  • Data models: Users, Projects, Manuscripts, ChatSessions, Comments, PublishedDocuments.

Challenges we ran into

  • AI response parsing: Gemini responses can be JSON, partial JSON, or HTML, so parsing and fallback logic is needed.
  • Editor diff review: generating a structured diff between HTML blocks and letting users accept/reject changes.
  • Reliable autosave: saving both chat sessions and manuscript content with debounce and error handling.
  • DB reliability: handling TLS/SSL and connection pooling for MongoDB in different environments.
  • Approval workflow: enforcing admin approvals and guarding routes with both middleware and API checks.

Accomplishments that we're proud of

  • Integrated AI copilot that can both chat and propose manuscript edits.
  • End-to-end pipeline from project creation → manuscript drafting → publishing.
  • A review workflow that visualizes AI edits and lets authors control changes.
  • Solid auth + admin approval system with role checks and approvals tooling.

What we learned

  • AI integrations need defensive parsing and strict schema validation to be stable.
  • Autosave is best handled with debounced, multi-endpoint persistence.
  • Document editing benefits from block-level diffing to make AI changes trustworthy.
  • Approval workflows reduce spam/abuse and give institutions better governance.

What's next for Sevivra

  • Real-time collaboration (presence + concurrent editing).
  • Version history and file management.
  • Data Analysis Dashboard: Fortunesheets empowered by Gemini.
  • Export to Word (UI mentions it; backend not implemented yet).
  • Peer-reviewer workflows with structured review assignments and status tracking.
  • Notifications and activity feed that reflect real data (not just UI placeholders).
  • Connection to Mendeley Collections for the assignment of relevant literature to each project.

Built With

Share this project:

Updates