Back to catalog
Resume Matcher
Apache-2.0Self-hosted, Apache-2.0 ATS tool that rewrites resumes to a job description locally with Ollama or your own LLM key, unlike Jobscan's paid cloud scanner.
Overview
Resume Matcher is a self-hosted tool that rewrites a resume against a specific job description instead of only scoring it for missing keywords. It parses both documents, scores the match with vector similarity, then hands the gap to an LLM you supply, local through Ollama or a cloud key, to actually edit the wording.
Where Resume Matcher fits
No commercial product does exactly this as a single self-hosted package; the closest neighbours are cloud ATS scanners like Jobscan and Teal. Both run as hosted SaaS, charge a monthly or per-scan fee past a free quota, and keep the resume and job postings on their own servers. Resume Matcher covers the same ground, keyword extraction, a match score, gap analysis, then rewrites the resume text itself instead of leaving you to work keywords back in by hand.
Running it changes who owns the infrastructure, not just the price. The Docker image bundles the FastAPI backend and Next.js frontend behind one port, and TinyDB writes straight to a JSON file instead of a separate database service, so there is nothing extra to provision. The tradeoff sits with the model: Jobscan's matching runs on infrastructure you never see, while Resume Matcher expects you to wire in Ollama for local inference or an API key for OpenAI, Anthropic, Gemini or DeepSeek, so whoever runs it also owns the uptime and per-resume cost of that model.
What it does well
- Master resume plus job description in, a vector-similarity match score and keyword gaps out, built on Qdrant rather than a plain string diff.
- The AI Tailor step rewrites resume sections to close the gap instead of just listing which keywords are missing.
- Cover letter generation and structured interview prep reuse the same tailored resume, so context does not get re-pasted into a second tool.
- Ollama support means the whole pipeline, parsing, matching, rewriting, can run without a resume leaving the host.
- LiteLLM abstracts the model layer, so switching from a local model to Claude Haiku 4.5 is a settings change.
- Official images ship for linux/amd64 and linux/arm64 on GHCR and Docker Hub, with a health-check endpoint and OpenAPI docs.
Limitations
- A roughly 240-second timeout on the tailoring request cuts off local LLM runs before they finish, with an unhelpful "signal was aborted without reason" error, and the length is not configurable (open GitHub issue).
