Your AI/ML research, on one deck. A responsive, multi-user web app that aggregates AI/ML papers (Latest / Trending / Famous), lets you star papers to read, and reads them in-app with scroll-resume and a "read up to here" highlight — synced across your phone and PC.

🚀 Try it live · 📺 Watch the demo

• One deduplicated corpus, three views — Latest (arXiv), Trending (Hugging Face upvotes + recency), Famous (citations). Papers are merged across sources by arXiv ID.
• Sources — arXiv, Hugging Face Daily Papers, Semantic Scholar, Google Scholar (owner-only, flagged). Papers With Code is included but disabled by default — its public API was retired in 2026.
• Star to read later — a personal library synced via your account.
• In-app reader — arXiv HTML (MathML-safe, sanitized) with block-level resume + highlight; a pdf.js fallback with page-level resume + highlight for PDF-only papers (e.g. older "famous" classics).
• Resume where you left off — your scroll position and a "I finished here" marker persist per paper, per user.

Prerequisites: Node 20+, pnpm, Docker (for local Supabase).

pnpm install

# 1. Start local Supabase (applies migrations in supabase/migrations)
npx supabase start

# 2. Configure env — copy the printed keys into .env.local
cp .env.example .env.local
#    NEXT_PUBLIC_SUPABASE_URL      -> http://127.0.0.1:54321
#    NEXT_PUBLIC_SUPABASE_ANON_KEY -> the publishable key from `supabase status`
#    SUPABASE_SERVICE_ROLE_KEY     -> the secret key from `supabase status`
#    OWNER_EMAILS                  -> your email (enables manual refresh + Scholar)
#    CRON_SECRET                   -> any string

# 3. Run the app
pnpm dev

# 4. Seed the corpus (pulls real papers from the live sources)
curl http://localhost:3000/api/cron/refresh -H "Authorization: Bearer <CRON_SECRET>"
#    ...or sign in as an OWNER_EMAILS user and click "Refresh" on /feed.

app/            routes (feed, paper, reader, library, login, api/*) + server actions
components/     UI: PaperCard, FeedTabs, reader (HtmlReader/PdfReader/ReaderBar), ...
lib/sources/    one adapter per source -> NormalizedPaper[]
lib/corpus/     dedupe + upsert + feed queries
lib/reader/     sanitize (MathML-safe) + fetch HTML + anchor/resume logic
lib/db/         Supabase clients (browser/server/service) + queries
supabase/       Postgres schema + RLS migration
docs/           design spec, implementation plan, DEPLOY.md

See docs/DEPLOY.md.

Contributions are welcome — especially new paper-source adapters (one file in lib/sources/). See CONTRIBUTING.md to get started, and SECURITY.md for reporting vulnerabilities privately.

MIT

⭐ If PaperDeck saves you a few tabs, consider starring the repo — it helps other researchers find it.