v0.2.2 — Admin Dashboard: LaTeX Inspector & GCP Log Viewer

What's New

🔬 LaTeX Inspector Tab

• Side-by-side viewer: Compare original English and translated Chinese LaTeX source code with synchronized line numbers
• File tree sidebar: Browse all .tex files for any paper with status indicators (original/translated)
• Paper selector: Searchable grid of all translated papers with translation status badges
• One-click download: Download the entire LaTeX package as .tar.gz archive for offline debugging
• Copy-to-clipboard: Quick copy for either original or translated panel content

📊 GCP Logs Tab

• Terminal-style viewer: Dark-themed log display with severity-coded badges (ERROR/WARNING/INFO)
• Service toggle: Switch between backend and frontend Cloud Run services
• Smart filtering: Filter by severity level, time range (1h to 7d), and result limit (50-500)
• Auto-refresh: Toggle 30-second polling for real-time monitoring
• Expandable messages: Collapse/expand long log entries for cleaner viewing

🛠️ Infrastructure

• Proxy route updated to stream application/gzip responses for tar downloads
• Backend admin endpoints for tex file listing, content retrieval, tar archive generation, and GCP Cloud Logging queries

Full Changelog: v0.2.1...v0.2.2

v0.2.1 — Cached Signed URLs, Simplified Naming, Polished Docs

🚀 ReadPaper v0.2.1

What is ReadPaper?

An AI-powered tool that translates arXiv papers from English to Chinese while preserving the original LaTeX layout — equations, citations, figures all stay pixel-perfect. Powered by Gemini 3.0 Flash with 1M-token context.

✨ Highlights in v0.2.1

⚡ Cached Signed URL PDF Delivery

• PDFs now served via GCS signed URLs — browser downloads directly from storage (1 hop, no backend proxy)
• Module-level credential caching + 12-min URL cache = near-instant responses on repeat access
• Fallback to direct streaming if signing fails

📛 Simplified PDF Naming

• Translated PDFs are now consistently named {arxiv_id}_zh.pdf
• Version suffixes (v1, v2) are stripped from output filename
• Single-candidate lookup (no more v1–v5 scanning)

📚 Polished Documentation

• README: attractive badges, live demo link, architecture diagram, feature highlights
• ARCHITECTURE: Mermaid flow diagrams, IPC protocol, signed URL delivery flow, storage layout

📊 Since v0.1

Metric	Value
Commits	11
Files changed	20+
Lines	+2500 / -800

Key Features

• 🧠 Whole-file LaTeX translation — one Gemini API call per .tex file
• 🔄 AI compile fix loop — auto-fixes LaTeX errors (3 retries)
• ⚡ GCS signed URL delivery — cached, 1-hop PDF loading
• 🔐 Google OAuth — per-user paper libraries
• 📊 Admin dashboard — user/paper management
• 📈 Live token tracking — real-time Gemini usage
• 💾 Translation cache — skip re-translation for cached papers
• 📖 Split-view reader — side-by-side EN/中文 with zoom & notes

v0.2: Admin Dashboard, Signed URL PDF Delivery & Cloud Run Optimization

🎉 ReadPaper v0.2

✨ New Features

• Admin Dashboard — Full admin panel at /admin with:

  • System overview (total users, papers, translations)
  • User management with paper/data deletion
  • Paper management with re-compilation support

• GCS Signed URL PDF Delivery — PDFs are now served via time-limited signed URLs directly from Google Cloud Storage, eliminating the backend proxy bottleneck that caused 5+ minute load times for 6MB PDFs.

• Translation Cache — GCS-backed cache with integrity validation. Previously translated papers are served instantly without re-using Gemini API credits.

🚀 Performance Improvements

• Cloud Run Optimization:

  • --cpu=2 for faster LaTeX compilation
  • --timeout=900 (15 min) for long papers
  • --no-cpu-throttling — background PDF compilation gets full CPU even after the HTTP request returns (fixes the root cause of compilation timeouts)

• Eliminated list_files() overhead — Backend now checks known PDF paths directly in GCS instead of listing all blobs under the prefix.

• Frontend proxy streaming — Binary responses (PDFs) are streamed through the Next.js proxy instead of buffered, preventing serverless function timeouts.

🔧 Bug Fixes

• Fixed cloudbuild-hotfix.yaml missing GOOGLE_CLIENT_ID and DISABLE_AUTH env vars
• Fixed _TaskStatusProxy.pop() missing method
• Fixed frontend PDF viewer cross-origin redirect failure with signed URLs

📚 Documentation

• Updated README.md with v0.2 features and Cloud Run configuration
• Updated ARCHITECTURE.md with PDF delivery flow, cache system, admin dashboard, and Cloud Run settings

Files Changed

10 files changed, 1229 insertions(+), 299 deletions(-)

v0.1: Stable Bilingual ArXiv Reader

ReadPaper v0.1 — First Stable Release

ReadPaper translates arXiv papers from English to Chinese while preserving full LaTeX layout, equations, citations, figures, and tables. Powered by Gemini 3.0 Flash (1M context window).

✨ Key Features

• Whole-file translation: Each .tex file → single Gemini API call (no text-node extraction, no batching)
• asyncio.gather concurrency: Parallel multi-file translation with configurable semaphore
• AI compile fix loop: latexmk -xelatex with up to 3 Gemini-powered error fix retries
• Dynamic compile timeout: 300s base + 60s per 10k output tokens (cap 1200s)
• Token usage tracking: Real-time Gemini API token display in frontend (In/Out breakdown)
• Concurrent translation guard: Prevents race conditions from duplicate submissions
• Split-view reader: Side-by-side bilingual PDF viewing (Next.js frontend)
• Cloud-native: Google Cloud Run + GCS deployment (cloudbuild.yaml)

✅ E2E Verified With

Paper	arXiv	Files	Total Tokens
DeepSeek-V3	2412.19437	4/4	15,923
Qwen2.5	2412.15115	9/9	69,803
GLM-5	2602.15763	11/11	65,262