TL;DR
We were building voice agents in healthcare and spending too much time testing them manually, and they still failed in production due to an edge case. Cekura came out of that experience. We help companies building AI voice agents deploy more reliable bots faster. We test efficiently, identify issues early, and demonstrate reliability before going live—all while enabling rapid development thereafter.

Try it yourself here or follow our demo here.

Problem: Anything can go wrong in production

Most teams building in voice, including us, have faced these challenges:

1. Demonstrating reliability to a client before going live
2. Manually calling & testing their AI voice agents, which is slow & error-prone.
3. Identifying all the edge cases in their workflow

Our Solution: Automated evaluations using adversarial voice agents

Run Monte-Carlo-like simulations using curated test sets. Here’s how Vocera helps:

1. Simulate Conversations: Our AI recreates realistic interactions using a library of workflows and personas.
2. Replay Old Conversations: Use real voice from painful production calls to run accurate simulations
3. Automated Scenario Generation: Pass in your call recording, and our AI will generate all the possible scenarios.

Our Ask

if you are building voice agents, we’d love to talk. If you know someone building voice agents, we’d still love to talk.

Email us at founders@vocera.ai. You can also book time with us here.

The Team:

We met over eight years ago during our undergraduate at IIT Bombay.

Tarush comes from quantitative finance, where he worked on simulations for ultra-low latency trading strategies (think nanoseconds!).

Shashij has previously researched NLP at Google Research and is the first author of a paper on testing AI systems reliably, which has 50+ citations from his work at ETH Zurich.

Sidhant comes from a consulting background advising CEOs at Fortune 500 companies in FMCG and medical devices. He managed P&L in a 10 Mn+ ARR in a conversational AI company.