ResumeForge

Local-first resume builder: a FastAPI backend renders JSON Resume data through themeable Jinja2 templates to HTML/PDF/PNG (headless Chromium), with a React editor UI, AI-assisted PDF import, and a server-side resume library.

Quick start

# Backend deps (one-time)
./scripts/setup.sh            # or: pip install -e . && playwright install chromium

# Frontend (one-time, then after UI changes)
cd web && npm install && npm run build && cd ..

# Run — serves the API and the built UI on one port
.venv/bin/python -m uvicorn src.main:app --port 8010
# open http://127.0.0.1:8010/

For frontend development with hot reload, run npm run dev in web/ (it proxies /api to port 8010).

PDF/PNG rendering uses Playwright's bundled Chromium and falls back to a system-installed Google Chrome when the bundled browser is unavailable.

Features

• Editor — schema-driven forms for all JSON Resume sections, plus a raw JSON escape hatch.
• Themes — themes/<name>/ each with a theme.toml manifest declaring design tokens (colors, fonts) and options (toggles/enums), all adjustable live in the UI:
  • red-twocolumn — serif two-column with accent header band
  • slate-modern — dark full-bleed sidebar, contemporary sans
  • ats-classic — single-column, system fonts, ATS-safe
  • executive-serif — centered classic serif
  • teal-banner — full-width banner header over two columns
  • minimalist-mono — Swiss minimal with left-gutter section labels
  • timeline-pro — vertical timeline experience
  • compact-tech — dense two-column for content-heavy resumes
• Font & color controls — per-theme font pickers (Google Fonts loaded on demand) and editable color tokens, with logical page breaks for clean PDFs.
• Live preview — debounced HTML/PDF preview; one-click export to PDF, PNG, or JSON.
• Persistence — the workspace autosaves to localStorage; named resumes save to resumes/<id>/resume.json via the library controls (save / load / delete).
• Variants — per-target overlays that reorder or hide sections (stored under x-meta).
• AI import — upload an existing resume PDF; a local LLM extracts structured data for review before committing.
• Profile photo — upload, auto-downscale, and an optional AI background blur (MediaPipe selfie segmentation, with an offline soft-focus fallback).

API

Method	Path	Purpose
GET	/api/themes	Theme manifests (tokens + options)
GET	/api/resumes	Library listing (id, name, label, updated_at)
POST	/api/resumes	Create/replace a resume
GET/PUT/DELETE	/api/resumes/{id}	Load / update / delete
POST	/api/render	Render {resume, theme, variant, format, tokens, options}
POST	/api/import/upload	AI extraction from an uploaded PDF

Development

.venv/bin/python -m pytest tests/ -q     # test suite
.venv/bin/python -m ruff check src tests # lint
cd web && npm run build                  # typecheck + bundle the UI

Adding a theme

Create themes/<name>/ with a theme.toml (name, engine, entrypoint, consumed sections, token defaults, options) and the Jinja2 template it points at. Tokens become CSS custom properties; options are available as options.* in the template. An optional style.css is rendered through Jinja and inlined into every output format. The theme appears in the UI automatically.

Configuration

AI import calls a local OpenAI-compatible LLM endpoint (e.g. llama.cpp, Ollama, vLLM). Defaults can be overridden with environment variables:

Variable	Default	Purpose
RESUMEFORGE_LLM_URL / OPENAI_BASE_URL	http://127.0.0.1:8080/v1	OpenAI-compatible base URL
RESUMEFORGE_LLM_MODEL / OPENAI_MODEL	qwen3.6-35B	Model name
RESUMEFORGE_LLM_THINKING	0	Set 1 to keep a reasoning model's chain-of-thought

The AI photo background blur downloads its segmentation model from a CDN on first use (cached afterward); offline, it falls back to a soft-focus blur.

License

GPLv3 © OpenSensor.