Case Study: Kastle ships compliance-grade voice AI for FDIC-insured banks with Cekura

Kastle ships compliance-grade voice AI with Cekura, cutting cost-per-call by 70%

Kastle, a fintech company providing voice AI agents to FDIC-insured banks, faced a significant challenge in safely iterating its graph-based agents. Due to strict regulatory requirements, every borrower call is auditable, and failures could result in compliance events. Off-the-shelf evaluation tools were insufficient because they provided only coarse, full-call assessments, making it impossible to pinpoint which specific state in an agent's graph had broken or failed a compliance obligation.

Cekura provided a node-based simulation and testing platform to address this. The solution allowed Kastle to build dedicated test suites for every individual state in its agents, running scenarios first in chat for broad coverage and then in voice for realism. This enabled compliance metrics like RESPA and NACHA to be scoped precisely to the node where each obligation lives. As a result, Kastle now ships agents with rigorous regression coverage, achieving a 70% lower cost-per-call, a 40% reduction in handle time, and over $90M in processed cash transactions while ensuring compliance-grade quality.

Kastle