KAVI Research Infrastructure: Building the Quality Standard for 3,000 Transcripts

Akshat joined KAVI as the fifth employee when the company was building a primary research transcript library from scratch. By the time he left, the team had grown to 45 people and the library held over 3,000 transcripts.

The core challenge was defining what a good transcript actually looked like. Quality in primary research is inherently subjective, and without a clear standard, output would vary across researchers and degrade as the team scaled. Akshat built the framework that solved this.

He defined quality along two measurable dimensions. The first was the number of substantive questions asked during a call: a transcript with too few questions indicated the researcher had not pushed deep enough. The second was answer length: when an expert shifted into storytelling mode and gave long, discursive answers, it was a signal the question had not been sharp enough to elicit a direct response. Akshat set a threshold, targeting expert answers under a certain word count as a proxy for precision.

Every transcript produced by the team, whether by Akshat or a colleague, passed through his review before moving to the content team for cleanup and publication. He was the quality gate. This model held as the team scaled, ensuring the library maintained a consistent standard across thousands of calls.