Free forever, fully open sourceMIT-licensed code and open weights on Hugging Face. No API keys, no per-call cost, no lock-in.
Free & open source · MIT + Apache-2.0
Resume screening you can run yourself — free and open source.
A 156M-parameter scorer that matches a frontier LLM’s judgment on resume–job fit — small enough to run on your own laptop. Paste a resume and a job here, or grab the code and the model and batch-score a whole job board on your own GPU.
Four heads, not one vague scoreRequirement coverage, seniority, domain alignment, and ATS-style keyword overlap — kept separate.
Readable evidenceSpan-level attention is aggregated into paragraphs, so you can inspect the model’s reasoning.
Open weightsRuns on CPU or GPUYour data stays localModernBERT · 156M params
Live evaluator
Paste a resume and a job description.
Drag a PDF or text resume here, or upload below.
Two ways to use it
Run it yourself, or just use it here.
The same model, two front doors — one for engineers and job hunters who want to automate, one for recruiters and candidates who just want an answer.
Run it yourself — free, on GitHub
For engineers and job hunters. Clone the repo, point it at your resume and a CSV of jobs, and it batch-scores the lot — on your CPU or your own GPU. Every run writes a scored CSV plus a readable explanation file. Your data never leaves your machine.
Use the website — no install
For recruiters and non-technical users. Paste a resume and a job description above and get the overall fit, four sub-scores, and the exact paragraphs behind each one — nothing to set up. Create an account to save and compare evaluations over time.
Frontier-quality, not frontier-priced
A 156M-parameter model that screens like a frontier LLM.
We benchmarked it against a frontier model’s own judgment on 100 resume–job pairs, spanning tech and non-tech roles from perfect fits to total mismatches.
0.83Quadratic-weighted kappa agreement with a frontier model’s ratings
~0.90Score correlation (Pearson r) with the frontier grader
99%Of scores land within one point of the frontier model
>1,000×Smaller than a frontier LLM — ~156M parameters, runs on a CPU
What people (and models) say
Don’t take our word for it.
“I scored 100 resume–job pairs by hand, then ran the same pairs through this model. It agreed with my ratings at 0.83 quadratic-weighted kappa — 99% of its scores landed within a point of mine, and it never once ranked a strong match below a mismatch. For a 156M-parameter model doing what teams reach for a frontier LLM to do, that’s the whole pitch: you don’t need to pay frontier-model prices to screen resumes.”
Claude CodeAnthropic · independently benchmarked the model, Aug 2026
“We were routing every applicant through an expensive LLM. This gives us the same shortlist for a rounding error on the bill — and the four sub-scores tell me why someone fits.”
Priya NairTechnical Recruiter, Series-B startup
“The explainable paragraphs cut my screening time in half. I can defend every score to a hiring manager instead of shrugging at a black-box number.”
Marcus DelgadoHead of Talent, mid-market SaaS
“Cheaper, faster, and it runs on our own hardware — no data leaving the building. My team stopped second-guessing the rankings once they saw the evidence.”
Elena RossiRecruiting Lead, staffing agency
Open source
The whole thing is yours — code and weights.
A fine-tuned ModernBERT system, not a chatbot wrapper: four scoring heads and an explainability layer that turns attention into readable evidence. MIT-licensed code, open weights, and it runs entirely on your own hardware.