Open-source · Not affiliated with Retell AI
Retell Assure catches failures
after launch.
Who catches them before?
VoiceIQ EVA is the missing pre-launch layer — EVA-Bench enterprise scenarios, EVA-A/EVA-X scoring, and a deploy gate for voice-agent prompts. Built to plug into the Retell ecosystem, not compete with it.
The Assure Gap
Without VoiceIQ EVA
- ✗Agent sounds great in demo
- ✗Ships to production
- ✗Healthcare auth loops on call #1
- ✗Assure flags it post-launch
- ✗Customer already churned
With VoiceIQ EVA
- ✓15 EVA scenarios before launch
- ✓Grounded backend + caller facts
- ✓EVA-A / EVA-X composite gate
- ✓Block deploy below 60% pass
- ✓Assure monitors clean production
Full quality loop
01 · BUILD
Write your Retell agent prompt
Domain LLM + tools + voice settings. You think you're done.
02 · VoiceIQ EVA
Pre-launch torture test
15 EVA-Bench enterprise scenarios. Simulated callers with real NPIs, PINs, ticket IDs. EVA-A + EVA-X scoring. Deploy gate.
03 · SHIP
Go live on Retell
Phone number live. Real humans calling. No undo button.
04 · Retell Assure
Post-launch QA
Monitor production calls. Catch regressions. The other half of the loop.
What broke — and what we fixed
Same model. Same prompts. Different grounding. 33% → 100% composite pass.
Before
Caller gives fake NPI. Agent loops. Everyone hangs up.
After · Healthcare HR
Caller facts injected from EVA database. Auth in 2 turns.
Before
Agent offers flight that doesn't exist. EVA-A: 0.12.
After · Airline CSM
Backend state + resolution script. 100% airline pass.
Before
Datadog call gets Wi-Fi troubleshooting script.
After · Enterprise ITSM
Scenario-specific ITSM scripts. Adversarial denials pass.
15
Scenarios
100%
Composite pass
97%
Avg EVA-A
3
Domains
For the Retell team
I built the pre-launch half of your QA story.
Open-source adapter for EVA-Bench. Live demo, certificate, HF Space, GitHub. Happy to walk through findings or contribute to your ecosystem.