Ertas for Resume Screening & HR

Fine-tune models that evaluate resumes against your specific role requirements, skills taxonomy, and hiring criteria, consistently and at scale.

The Challenge

Talent acquisition teams at growing companies review hundreds or thousands of applications per open role. Manual resume screening is time-consuming, inconsistent, and prone to unconscious bias. Different recruiters apply different standards to the same role, qualified candidates are overlooked during high-volume periods, and the pressure to move quickly leads to superficial reviews that miss transferable skills or non-traditional backgrounds.

Existing AI screening tools use keyword matching that is easily gamed and fails to evaluate actual qualification. A candidate who lists every technology keyword in their resume scores higher than a genuinely skilled engineer with a concise resume. These tools also cannot adapt to organization-specific requirements. Your definition of 'senior' may differ from the generic industry standard, and the specific technology stack, domain experience, and soft skills that predict success at your company are unique to your hiring context. Off-the-shelf screening tools optimize for generic job descriptions, not for your specific team's needs.

The Solution

Ertas lets HR teams fine-tune screening models on their organization's own hiring data: resumes that experienced recruiters evaluated, paired with structured assessments written against the team's rubric. The model learns which qualifications, experience patterns and skill combinations the organization looks for in each role type, and reads what a candidate's experience means for the role instead of counting keywords.

In Ertas Studio, talent teams train on pairs of resumes and structured evaluations that rate each candidate on technical fit, experience relevance, skill alignment and other dimensions the hiring team defines. The fine-tuned model then reads new applications and writes assessments in the format recruiters already use. Exported from Hub and run on the company's own servers, it can sit behind the applicant tracking system as a first-pass prioritization step, with a person still reviewing every candidate. Training runs on Ertas's cloud GPUs, so the training set is uploaded to Ertas: remove names, contact details and demographic information first. Meeting employment and data protection law, telling candidates about automated screening and testing the model for bias remain the employer's own work.

Key Features

Studio

Custom Evaluation Criteria Training

Train screening models on your evaluation rubrics, role requirements and recruiter assessments in Studio. The model learns your organization's own definition of a qualified candidate for each role.

Studio

Rubric Examples in Data Craft

Build and review training examples in Data Craft, Studio's guided dataset builder. Senior recruiters edit the target assessments, version them and approve each one, so the model learns from evaluations the team agrees with.

Hub

Screening Model Behind Your ATS

Hub lists your trained models. Download one as GGUF or full safetensors, serve it with Ollama or vLLM, and call it from your ATS integration or an n8n workflow to return structured evaluations into the recruiter queue.

Hub

Candidate Data Stays In-House at Screening

After export the model runs on your own servers, so the resumes it screens and the evaluations it writes stay inside your network. Training data is uploaded to Ertas, so de-identify it before upload, and delete datasets from Studio when your retention period ends.

Example Workflow

A technology company hiring 200 engineers a year receives about 500 applications per role. The talent team exports 20,000 historical applications with recruiter evaluations on a five-dimension rubric, strips names, contact details and demographic identifiers, and uploads the set to Ertas Studio. Senior recruiters review a sample in Data Craft and correct evaluations they disagree with. The team fine-tunes a screening model, compares runs on a held-out set of applications the model never saw, and runs its own bias checks before go-live. The chosen model is exported and served on the company's infrastructure behind the ATS. When new applications arrive, the model writes a structured evaluation against the recruiter rubric, with a priority score and a short rationale citing specific resume elements. Recruiters still review every candidate and start from the prioritized list. The team measures screening time and interview rates against its pre-model baseline and retrains when rubrics change.