PaperBanana is an AI illustration generator for academic work, aiming to get the figures in papers, reports and course slides usable faster.
I lead the core development — Codex as the primary implementer, Claude assisting with architecture decisions — turning a research-illustration idea into a deliverable multi-platform system.
Scope of delivery
- Shipped five clients from a single codebase: Web, macOS, Windows, Android and a WeChat mini-program.
- Organized the monorepo with pnpm workspaces, sharing API calls, type contracts and product logic across packages.
- Rewrote the upstream Python prototype into a zero-dependency TypeScript CLI published to npm, collapsing “set up a Python environment + multi-step install” into a single command.
- Deployed a Better Auth gateway and Laf cloud functions on Sealos for cross-platform account, task and generation-history sync.
Current progress
The production site paperbanana.asia is live. Next up: more real research-scenario templates, a better editing flow for generated results, and a technical write-up spanning model adaptation to multi-platform release.