Custom Voice AI for Screening Entry-Level Applicants

Recruiting teams often start with automated interview platforms like Spark Hire or Willo. These tools are great for replacing one-way video screens but fail at accurately assessing technical roles. Their scoring models are black boxes, often grading for generic traits like 'clarity' which has little correlation with a developer's ability to write Python code.

A specialized recruiting firm tried using one of these platforms to screen for 10 different engineering roles. The platform could not differentiate a good answer about AWS services from a bad one. Recruiters had to listen to every 15-minute recording to manually score technical competence, which completely defeated the purpose of automation. They were paying $120 per recruiter per month for a glorified audio file host.

Furthermore, these platforms offer limited integrations. They might push a candidate's overall score to a major ATS like Greenhouse, but they cannot sync detailed, per-question feedback to a custom field or connect to an industry-specific recruiting system. This forces a workflow of manual copy-pasting that introduces errors and wastes time.

Syntora's approach would begin with a discovery phase to meticulously map your existing screening questions and scoring rubric. This translates into a set of precise prompts for the Claude API, engineered to reflect your specific evaluation criteria for various roles.

The technical architecture would involve a Python and FastAPI endpoint, designed to accept audio files submitted from your careers page or application form. This endpoint would be deployed as a serverless function on AWS Lambda, allowing for scalable handling of unpredictable applicant volume without incurring continuous server costs.

Upon receiving an audio file, the Lambda function would first route it to a transcription API, such as Deepgram. Deepgram is chosen for its robust performance and high accuracy, particularly with technical jargon commonly found in specialized roles like junior developers. For instance, in other document processing pipelines we've built using Claude API (for financial documents), similar transcription services consistently achieve high word accuracy, a pattern that applies to technical screening scenarios. A typical 3-minute audio response can be transcribed into text rapidly, usually in under 4 seconds.

The generated transcript is then sent to the Claude 3 Sonnet API. The custom-designed prompt would instruct the model to score the candidate's answers against the defined rubric, providing ratings from 1 to 10 across criteria such as technical correctness, problem-solving approach, and cultural fit. The Claude API is designed to return a structured JSON object, including scores, a summary, and direct quotes supporting the evaluation, typically within about 9 seconds.

Finally, this structured JSON output would be stored in a Supabase database for historical analysis and reporting. Concurrently, an asynchronous function would make an API call to your applicant tracking system (ATS) to populate custom fields with the per-question scores and summary. The client would need to provide access to their ATS API and define the specific fields for integration. The delivered system would provide the core logic for voice-based applicant screening, with clear APIs for integration into existing career portals and ATS systems. An engagement would include architectural design, API development, cloud deployment, and comprehensive documentation, enabling your team to maintain and extend the system. Typical end-to-end processing for a 3-minute audio file, from submission to ATS update, can be expected to complete within 15 seconds, with hosting costs for the serverless components often under $50 per month for hundreds of applicants, depending on actual usage.