Rushil Kumar
Get in touch Resume
← All Work

TruthSeek · live at truthseek.in

A voice-based user research platform: it runs interviews and turns the conversations into research a team can actually use, not just raw transcripts.

100xinterview throughput over manual research
30 minto run 100 interviews in parallel, versus 10 hours for 20 manual calls

The problem

User interviews are useful, but organizing, running, and analyzing them by hand doesn't scale past a handful of calls. The hard part isn't automating the call itself. The agent has to ask the right questions, respond naturally, stay inside the research brief, and produce something a researcher can actually use, not just a transcript to dig through.

What I built

  • Human-in-the-loop live-call monitoring. Built the feature that lets a researcher watch an AI calling-agent run an interview in real time and take over the call mid-interview if needed.
  • Calling-agent guardrails. Engineered the guardrail layer governing question wording, company-mention handling, intent exposure, and follow-up depth, so automated interviews stay compliant and on-script without a human writing every question by hand.
  • Insights dashboard integration. Integrated Insights Engine, originally a standalone review-intelligence project, as a real-time analytics dashboard inside TruthSeek, giving product-growth teams at FMCG brands live visibility into the consumer insights gathered from their interviews.

Outcome

TruthSeek moved from an early idea to a working product that multiple customer teams now use. Across every interview format, a capable model alone was never enough. It also took a clear interview structure, explicit behavior controls, reliable transcript processing, and output a researcher could actually trust.

Stack

Python · voice-agent workflows · LLM guardrails · real-time dashboard integration