CV Tailor (local-first)

Polski / Polish version → README.pl.md

Local-first, privacy-first app for honestly tailoring your CV to specific job offers. It does not invent experience — it highlights what you actually have and shows gaps as suggestions.

Stage: v1.0 — stable feature set. Full flow: import CV → job offer → analysis → tailored CV → optional cover letter → export to Markdown / PDF. PL/EN UI. Your data and API key stay local.

Features

• Paste or load a CV (.txt, .md, .docx) and a job offer.
• Split the CV into sections (summary, experience, skills, education, and detected custom sections).
• Extract requirements from the offer (hard / soft skills).
• Weighted matching with a percentage score (hard skills count double), plus matched / missing breakdown.
• Review-first tailoring: accept / reject individual suggestions with evidence mapping.
• Generate a tailored CV (offline heuristic or AI model — BYOK).
• Optional cover-letter generator (offline heuristic + AI variant, 3 tones).
• PL / EN interface (switch in the header or Settings).
• Project versioning (save / open / delete whole CV + offer projects) and saved CV versions.
• Multiple PDF layouts (classic / compact / modern).
• Local autosave; export to Markdown or PDF.

Getting started

npm install
npm run dev

The app starts on http://localhost:5173.

Other commands:

npm run build      # production build
npm run typecheck  # type checking
npm run smoke      # quick core-logic smoke test (no UI)
npm test           # unit tests (Node built-in runner, no extra deps)

AI (BYOK — Bring Your Own Key)

In the Settings tab you choose a provider and model. Two modes:

• Cloud providers (OpenAI, Groq, OpenRouter, Together, Mistral) require an API key.
• Local providers (LocalAI, Ollama, or a custom local endpoint) work without any key — the app simply omits the Authorization header.
• No AI at all: leave it unconfigured and the app runs fully offline using the heuristic generator.

The key (when used) is stored only in your browser's localStorage.

Example: running locally with Ollama

ollama serve
ollama pull llama3.1
# In Settings: provider "Ollama", model "llama3.1", no key needed.

Privacy

• Your CV, the offer and settings are never sent to any app server (there is no backend).
• The only outbound network request is the optional call to the AI provider you choose.

Documentation

• docs/architecture.md — layers and design decisions.
• docs/usage.md — step-by-step usage guide.
• docs/contributing.md — how to develop and test.
• docs/roadmap.md — milestones.

The core (src/lib/core) is independent of the UI and the browser — it is unit-tested and can later be moved into a separate package / sidecar (e.g. for the planned Tauri desktop build).

License

MIT — see LICENSE.md.