interlinear

A post-editing IDE for AI translations of PDFs. Translate a document page-by-page, then read and refine it side-by-side with the original.

The viewer renders each page as left = original PDF image, right = the translated page (a React/TSX component), so you always see the source and your translation in visual correspondence. There is no LLM runtime of its own — translation and editing are driven by Claude Code skills that run inside your own session.

• Each source PDF becomes a doc under apps/demo/docs/<docId>/, with one pages/page-NNNN/index.tsx component and one page-NNNN.png per page.
• A Vite plugin scans those pages, serves the viewer, and exposes edit endpoints so changes save straight back to the TSX files.
• Doc content stays out of git by default (only docs/example/ is committed).

pnpm install
pnpm dev        # builds core, then starts the demo viewer (default :5173)

Open the printed URL to browse the example doc.

pnpm --filter @interlinear/demo build   # → apps/demo/dist/

Deploy apps/demo/dist/ to any static host (Cloudflare Pages, Netlify, GitHub Pages, S3+CDN) — there is no backend or database. The build:

• renders every page to static JS and copies each doc's public/ images,
• emits a per-doc search-index.json, so the reader's full-text search (across both the translation and the original PDF text) runs entirely client-side.

The authoring tools (inline editing, DocChat) are dev-only and don't ship in the static build. Cross-language search over the original text uses the .interlinear/original-text.json cache — open search once under pnpm dev (or run the text extractor) before building so that cache exists; without it the published index still covers the translated text.

Run these from a Claude Code session in this repo:

packages/core   framework: React runtime, Vite plugin, skills
apps/demo       the viewer app and doc workspace