TabScribe

Research OS for the Web

💡 Inspiration

Research today is chaos.

Students, scientists, and writers constantly jump between tabs — summarizing on Gemini, translating with Google Translate, rewriting in ChatGPT, and copying everything back into Docs or Notion. I faced the same pain every single day while preparing research notes and citations.

That's when I realized: "Why not make the browser itself smart enough to do it all?"

That question became TabScribe — a Research OS for the Web, built entirely on Chrome's Built-in AI (Gemini Nano).

✨ What It Does

TabScribe turns your web browsing into a structured research workspace. With one click, it can:

Capture text, metadata, and evidence snapshots from any page
Process with summarize, rewrite, translate, or proofread using Gemini Nano
Organize all snippets into clean, searchable research cards
Export polished drafts with citations in APA, MLA, or BibTeX
Discover literature optionally via Hybrid Mode with OpenAlex & Crossref

It's fully offline-first, private by design, and runs entirely on-device.

🛠️ How I Built It

Framework: Manifest V3 Chrome Extension
AI Engine: Summarizer, Rewriter, Translator, Proofreader, and Writer APIs from Chrome's Built-in AI
Storage: IndexedDB + Chrome Storage for offline data persistence
Frontend: HTML, CSS, and Vanilla JavaScript for lightweight performance
Visualization: D3.js for citation network graphs
Hybrid Mode: Gemini API for literature metadata integration

🚧 Challenges I Faced

The early Chrome AI APIs were experimental, which meant debugging response timing and local permission models took weeks of iteration. Synchronizing IndexedDB between the sidebar and background scripts introduced unexpected complexity. I also had to carefully balance offline privacy with optional hybrid connectivity—getting the UX signaling right was crucial.

The biggest challenge was compressing multimodal features (image + audio + text) into a lightweight extension while keeping performance fast and snappy.

🎓 What I Learned

How to work directly with Chrome's on-device AI stack and handle multimodal prompts efficiently
The power of designing AI experiences that run locally, not in the cloud
Building small but expressive UIs that make AI features feel native inside the browser