Kovi
An AI interview assessment system built on LangGraph. It uses isolated evaluation nodes to decouple scoring from dialogue, and is designed to automatically evaluate back-end engineering candidates.
Timeline (last 90 days)
The Kovi team conducted a double-blind benchmark test, having an AI evaluation engine based on LangGraph score and compare 25 anonymized technical interview records against three senior Staff Engineers.
Across all 25 interviews, the Kovi AI evaluator achieved an overall correlation of 94.2% with the human baseline.
In the composite score, the human average was 7.37 and the Kovi average was 7.25, an overall gap of only 0.12 points.
On the system design dimension, the human average and the Kovi average were both 7.1, completely identical.
On the communication dimension, the human reviewers' average was 8.2 and the Kovi average was 7.8, with Kovi being stricter (deviation -0.4).
In all 25 evaluations, Kovi never cited technologies or frameworks that the candidate had not explicitly mentioned, i.e. zero scoring hallucination.
Kovi uses an Isolated Evaluator Node, with the evaluation model completely decoupled from the conversational speech model, and the evaluator's input strictly limited to the interview transcript text.
Kovi's cost per screening is 150 rupees, about 13 yuan or 1.8 US dollars.
An AI evaluation system based on a deterministic state machine can run thousands of evaluations simultaneously without degradation, with constant marginal cost, whereas human reviewers become fatigued and their attention declines over time, with increasing unit cost.
All Facts (9)
An AI evaluation system based on a deterministic state machine can run thousands of evaluations simultaneously without degradation, with constant marginal cost, whereas human reviewers become fatigued and their attention declines over time, with increasing unit cost.
50%UnverifiedAcross all 25 interviews, the Kovi AI evaluator achieved an overall correlation of 94.2% with the human baseline.
50%UnverifiedIn the composite score, the human average was 7.37 and the Kovi average was 7.25, an overall gap of only 0.12 points.
50%UnverifiedOn the system design dimension, the human average and the Kovi average were both 7.1, completely identical.
50%UnverifiedOn the communication dimension, the human reviewers' average was 8.2 and the Kovi average was 7.8, with Kovi being stricter (deviation -0.4).
50%UnverifiedIn all 25 evaluations, Kovi never cited technologies or frameworks that the candidate had not explicitly mentioned, i.e. zero scoring hallucination.
50%UnverifiedKovi uses an Isolated Evaluator Node, with the evaluation model completely decoupled from the conversational speech model, and the evaluator's input strictly limited to the interview transcript text.
50%UnverifiedKovi's cost per screening is 150 rupees, about 13 yuan or 1.8 US dollars.
50%UnverifiedThe Kovi team conducted a double-blind benchmark test, having an AI evaluation engine based on LangGraph score and compare 25 anonymized technical interview records against three senior Staff Engineers.
50%