The problem
AccioJob's assessment funnel had a 50% false-positive rate: candidates who passed automated coding tests but failed live technical interviews. Every false positive cost recruiter time, client trust, and interview slots. The root cause: static assessments can be gamed, and typing-based tests don't verify understanding.
The build: 48 hours to production
We added a conversational AI voice interviewer as a verification layer after the coding round. Candidates explain their reasoning out loud; the agent probes with adaptive follow-ups, the 'why?' chain that's nearly impossible to fake. Neeraj integrated Retell AI's voice stack into the existing event-driven assessment pipeline in 48 hours, including scoring rubrics and recruiter-facing evidence reports.
The result
False positives fell from 50% to 15%, a 70% reduction, published by Retell AI as an official case study. Later, we re-architected the voice layer onto our own LiveKit stack, cutting per-minute cost from ~10¢ to 2.5¢ while preserving assessment quality.