Resume Builder
A self-hosted tool that turns a job description into a tailored, one-page PDF
resume. Paste a JD, let it pick the best-fit resume from your library and
suggest surgical bullet edits, then render a clean ATS-friendly PDF. It can also
draft a resume-grounded cover letter.
You bring your own AI key (Anthropic or OpenAI), or point it at a local model.
Nothing is sent anywhere except the provider you choose.
Scope: this is a local / single-user tool. It is not a multi-tenant SaaS —
resume isolation is keyed off your API key as a convenience, not a hard
security boundary. Run it for yourself, on your machine or your own server.
Paste a job description and it scores the fit, lists the keywords you match and the ones you lack, and recommends
which resume in your library to start from. The example uses invented content.
Features
- JD evaluation — fit assessment, role match, and keyword gaps.
- Auto-triage — picks the best resume from your library for the role.
- Surgical bullet edits — suggestions you approve before they land; never a
full rewrite, never more than one page.
- One-page PDF — Times New Roman, ATS-friendly reading order, via headless
Chromium.
- Cover letters — grounded in your own resume content.
- Bring your own model — Anthropic, OpenAI, or any OpenAI-compatible local
server (Ollama, LM Studio, vLLM).
Quickstart (Docker — recommended)
Docker handles the trickiest dependency: headless Chromium plus fonts.
cp .env.example .env # can stay blank for bring-your-own-key use
docker compose up --build
Open http://localhost:5001, click the gear, and add your API key (or a local
endpoint). Upload a resume .docx in Settings, paste a JD, and build.
Quickstart (local Python)
pip install -r requirements.txt
playwright install chromium
python app.py # http://localhost:5001
Using a model
Open Settings (gear icon) and choose one:
- Anthropic / OpenAI — paste your key (
sk-ant-... or sk-...). It is sent
with each request from your browser and never stored on the server.
- Local model — requires
ALLOW_CUSTOM_BASE_URL=1 (see below). Leave the key blank and enter an OpenAI-compatible endpoint,
e.g. http://localhost:11434/v1 for Ollama, plus a model name like
llama3.1. Runs fully offline. Larger models give more reliable structured
output; small ones may occasionally need a retry.
Get a key: Anthropic Console ·
OpenAI Platform.
Self-hosting a local model
Custom endpoints are off by default: the server refuses any request that
carries a base_url (HTTP 400) and the Settings panel hides the field. A
client-chosen URL would let any visitor make the server call addresses on its
own network, so this is only safe when you are the only user. If you run Resume
Builder on your own machine and want a local model, set ALLOW_CUSTOM_BASE_URL=1
in .env (or in the environment) and restart. Do not set it on an instance that
other people can reach.
Your resume data
- Upload resumes through the Settings panel (
.docx). They are parsed and stored
in data/resumes/ as JSON, tagged to your key so only you see them.
- A fictional sample resume ships so the app works out of the box. Delete it
once you've added your own.
data/, outputs/, .env, and *.docx are gitignored — your content stays
local and is never committed.
Owner mode (optional, for a personalized deployment)
If you host this for yourself, you can set OWNER_ACCESS_CODE in .env. Typing
that code into the API-key field swaps in the server's ANTHROPIC_API_KEY and
unlocks your private resumes — so the public sees a generic, bring-your-own-key
tool while you get your full setup on your own tokens. Owner mode is
rate-limited (OWNER_RATE_MAX / OWNER_RATE_WINDOW) so an exposed endpoint
can't drain your tokens. Leave the code blank to disable owner mode entirely.
How it works
- Flask backend (
app.py), Playwright + Chromium for PDF rendering
(resume_maker.py), python-docx for parsing uploads (resume_parser.py).
- AI is used only for decisions (triage, edit suggestions, summaries, cover-letter
prose). PDF generation itself makes zero AI calls.
- Cover-letter tone/structure templates live in
cover_letter_kb.yaml — copy
cover_letter_kb.example.yaml to create your own.
License
MIT — see LICENSE.
