Using Resurface in hiring

Resurface helps people read resumes faster. It does not decide who gets a job. This page explains what the service does, what stays your responsibility as the employer, and what we can do for your legal, security or procurement review.

You, as the employer, are responsible for the hiring and AI laws that apply where you and your candidates are. We are a small, bootstrapped product: if our cloud doesn't fit, you can run the open-source version yourself, and we'll help.

Who we are

Resurface is a bootstrapped product, built and run by its founder in Israel. There are no investors, no sales team and no account managers. That has costs we'd rather tell you about up front:

• We are not SOC 2 or ISO 27001 certified yet, and we don't offer an uptime SLA.
• One person handles support, security and incidents.
• Our providers are established companies with their own certifications, listed on the subprocessors page.

It also means you talk directly to the person who wrote the code. Questions, a security questionnaire or a contract review go to office@bulldog-adv.com, and we answer them ourselves.

What Resurface does

• An LLM drafts 6–8 screening questions from the job description, once per job. You can send up to 12 questions of your own instead, so you control what is asked.
• A classification model answers each question for each resume: SUPPORTED, CONTRADICTED or INSUFFICIENT_EVIDENCE, with the resume line that proves it. The line is copied from the resume, not written by a model.
• INSUFFICIENT_EVIDENCE means the resume doesn't say. It is not a failing answer.
• The 0–100 score summarises the answers. Every response names the model that answered, so a result can be traced.

What Resurface doesn't do

• It doesn't reject, shortlist or rank candidates. Your process decides what to do with a score.
• It doesn't look at photos, video or voice, and it doesn't infer age, gender, ethnicity, religion, health or any other protected trait.
• Its questions come from the job's requirements. Our terms forbid using it to discriminate on protected characteristics, and our DPA asks you not to submit special categories of data where you can avoid it.
• It doesn't train models on your resumes, and neither do our model providers.

What stays your responsibility

Laws on AI in hiring usually place their duties on the employer that uses a tool, not only on the tool's vendor. Depending on where you and your candidates are, they can require you to:

• tell candidates that an automated tool helps review their application, and what it looks at;
• have the tool's results checked for bias by an independent auditor and publish a summary (for example, New York City's Local Law 144 for candidates in the city);
• keep a person in charge of each decision, and let candidates ask for a human review;
• assess the risk of the tool before you use it.

Similar rules exist or are coming in other US states, in the EU (the AI Act treats AI used in recruitment as high-risk) and elsewhere. We are not your lawyers and can't tell you which apply to you. Under our terms, you are responsible for using Resurface lawfully and for the decisions you make with it.

What we can give you

• A written description of how a score is produced, for your notices to candidates or your risk assessment.
• Our model eval: before a decision model serves scores, its verdicts must agree with Claude Sonnet 5.5 on our resume set. The method and the benchmark code are open source.
• Help preparing data for a bias audit, such as exporting your results.
• Answers to your security questionnaire, our DPA, and the details in our trust center.

If our cloud doesn't fit

Some organisations can't send resumes to a small vendor, or to providers in the United States. For them, the open-source version of Resurface (MIT licence) runs on your own infrastructure:

• with your own TypeSafe and LLM keys, under your own contracts with those providers; or
• fully offline, with no model at all. In that mode it uses a keyword method that runs on your servers and sends nothing out. Its results are noticeably weaker than the cloud's, and it is labelled as such everywhere.

Either way, we'll make every effort to help your team set it up and get the most out of it. Write to office@bulldog-adv.com.