CUSTOMERS / NURIX
How Nurix Delivers Enterprise Grade Reliability for High Volume Voice AI Workflows
At the end of every call is a real person who just wants to be understood. Whether they're mixing languages or giving a quick one-word answer, our AI needs to listen with a human-like intuition. Partnering with Cekura allowed us to move beyond vibe based tests and stress-test our system against the messy, beautiful reality of human conversation ensuring that no matter how someone speaks, they feel heard.
Nurix
Company
About
Nurix's vision is to revolutionise customer engagement with intelligent, adaptive AI solutions that integrate deeply into enterprises' existing workflows seamlessly. By providing custom AI solutions tailored to enterprise needs, Nurix augments their workflows instead of disrupting them.
Industry
Enterprise AI
Company size
50-200
Key features
- • Voice AI for Sales and Customer Support
- • Workflow Automation
The Challenge
Nurix provides critical AI infrastructure for large enterprises, handling millions of high-stakes customer interactions. When deploying a flagship voice agent for a major enterprise client, they encountered three systemic hurdles that standard testing could not replicate:
Linguistic Complexity: The end-users often spoke in "Hinglish," mixing Hindi and English in the same sentence. Nurix needed to prove their agent could handle complex entity extraction (like specific product names or account types) without hallucinating during these language shifts.
Edge Case Robustness: The agent was fragile when handling "low-data" responses. User behaviors like responding with a single word or pausing for two seconds were triggering incorrect timeouts, causing the system to disconnect valid calls.
Concurrency at Scale: The deployment needed to support hundreds of simultaneous concurrent calls during peak operational windows. Early load tests revealed a 70% failure rate with latency drifting over 2.6 seconds, risking a complete service outage.
The Solution
To harden their infrastructure against these edge cases, Nurix replaced manual testing with Cekura's high-velocity, persona-driven simulations. The team launched multilingual "chaos testing" to stress-test the agent against complex mixed-language inputs. Simultaneously, they ran "Silent Customer" and "Short Answer" scenarios to identify and recalibrate aggressive Voice Activity Detection (VAD) thresholds. Finally, dedicated load benchmarking to enable the engineering team to optimize query paths prior to launch.
The Results
By integrating automated simulation into their CI/CD pipeline, Nurix hardened their AI infrastructure to meet strict enterprise SLAs:
1. Multilingual Validation: While initial mixed-language tests saw less than 70% success rate, the simulation data allowed Nurix to retune their model. They successfully patched the logic, ensuring seamless transitions between languages without losing user intent.
2. Elimination of "Short Answer" Drops: "False Silence" bug where the agent disconnected on users who gave concise one-word answers was identified and fixed. This single optimization could save thousands of valid interactions per month that would have otherwise been abandoned.
3. Stability Under Load: Stress testing exposed that 30% of calls failed at peak load due to timeouts. By catching this pre-production, Nurix optimized their webhook response times, bringing average latency down from 2.65s to sub-second levels, ensuring a snappy, real-time experience even during traffic surges.