01 / Problem & ownership
The product context and my responsibility.
- Product context
- Scientific authoring, compiled review, and AI assistance inside one collaborative workspace.
- Constraint
- Keep source, compile feedback, document preview, and assistant state understandable together.
- My ownership
- Frontend ownership across authoring, preview, AI chat, analytics, and admin tooling.
- Key decision
- Connect source, compiled feedback, PDF review, and AI actions in one product workflow.
- Primary outcome
- Scientists can edit source, review compiled output, and act on AI feedback without switching tools.
02 / Implementation
How the workflow came together.
- Authoring
- Editor shortcuts, autocomplete, minimap controls, auxiliary TeX files, plain-text fallback, and reliable saves.
- Compiled output
- PDF preview for LaTeX and Typst with compile errors mapped back to source and graceful fallback states.
- AI workflow
- Per-message model selection, reasoning blocks, token labels, inline errors, and reusable response actions.
- Product operations
- Amplitude instrumentation, environment-aware runtime config, paginated admin data, and EN/RU localization.
03 / Result
Scientists can edit source, review compiled output, and act on AI feedback without switching tools.
Current client engagement. This case describes delivered interface capabilities; no adoption or business-impact metrics are claimed.
Technology used
- React 19
- TypeScript
- Vite
- Zustand
- Monaco
- PDF.js
- Yjs
- Amplitude