PaperLens

PaperLens turns dense arXiv papers into a structured learning experience.
Instead of reading 20+ pages linearly, users get guided explanations, visual diagrams, concept cards, and an actionable learning path in one flow.

What users get

• Layered explanations (Simple, Engineer, Deep Technical)
• TL;DR plus a full summary suite (ultra-short, 1-minute, 5-minute, detailed)
• Concept cards with intuition, technical definitions, pitfalls, and mini quizzes
• Visual section with guaranteed output:
• 2 Mermaid flowcharts
• 2 custom infographic-style diagrams
• Section-by-section paper breakdown
• Learning path with step goals and estimated time
• Why-it-matters panel with use-cases and adoption context
• Light and dark themes with responsive UI across desktop and mobile
• Authentication: email/password + optional Google OAuth
• Personalized dashboard with recent papers, bookmarks, and related recommendations

Input model

• Accepted input: arXiv URL or arXiv paper ID
• No PDF uploads from users
• Gemini API key is server-managed (GEMINI_API_KEY)

Tech stack

• Next.js 16 (App Router, TypeScript)
• React 19
• Gemini (@google/generative-ai)
• Mermaid
• Framer Motion
• Lucide Icons
• NextAuth (JWT sessions)
• SQLite (better-sqlite3) for user auth + personalization state

Local setup

1. Install dependencies:

npm install

2. Create env file:

cp .env.example .env.local

3. Add required keys to .env.local:

GEMINI_API_KEY=your_key_here
DATABASE_PATH=data/paperlens.db
NEXTAUTH_URL=http://localhost:3000
NEXTAUTH_SECRET=replace_with_a_strong_random_secret

# Optional for Google login
GOOGLE_CLIENT_ID=
GOOGLE_CLIENT_SECRET=
# AUTH_GOOGLE_ID=
# AUTH_GOOGLE_SECRET=
# AUTH_SECRET=

4. Run the app:

npm run dev -- --port 3000 --webpack

Open http://localhost:3000.

Quality checks

npm run lint
npm run typecheck
npm run build

Or all at once:

npm run check

Production hardening already included

• Strict input validation for /api/analyze
• Graceful handling of malformed upstream/model JSON
• Fallback content path when PDF extraction fails
• no-store upstream fetch behavior for fresh paper data
• Security response headers (X-Frame-Options, Referrer-Policy, etc.)
• nodejs runtime enforced for analysis route
• Production-safe theme persistence and hydration

Project structure

src/
  app/
    api/analyze/route.ts      # analysis endpoint
    api/auth/*                # login/signup/session APIs
    api/user/*                # recent/bookmark/dashboard APIs
    auth/page.tsx             # login/signup UI
    dashboard/page.tsx        # personalized dashboard UI
    page.tsx                  # landing + orchestration
  components/
    HeroInput.tsx
    ThemeToggle.tsx
    dashboard/                # all learning sections
  lib/
    arxiv.ts                  # id parsing + metadata + text extraction
    gemini.ts                 # prompting + normalization + fallbacks
    types.ts

Deploy notes

• Recommended Node.js: >=20.11.0
• Set GEMINI_API_KEY, DATABASE_PATH, NEXTAUTH_URL, and NEXTAUTH_SECRET in your deployment environment
• Add GOOGLE_CLIENT_ID + GOOGLE_CLIENT_SECRET (or AUTH_GOOGLE_ID + AUTH_GOOGLE_SECRET) if Google OAuth is enabled
• Run npm run build in CI before release

---

If you are working on the product direction: prioritize reliability of parsing + clarity of learning output over adding new surface features. That is where user trust is won.