How to Create AI Judge Metrics You Can Trust
How to use human review in Coval to calibrate your LLM-as-a-judge metrics: label conversations, check agreement, read disagreements, fix the prompt, and re-test.
Coval's observations on developing and shipping AI voice agents
How to use human review in Coval to calibrate your LLM-as-a-judge metrics: label conversations, check agreement, read disagreements, fix the prompt, and re-test.
Voice AI observability captures every production call and judges whether the agent did its job. Here's what it captures, what to measure, and how failures become tests.
Vapi, Retell, LiveKit, Sierra, LangGraph and coding agents solve different jobs. Build a shortlist around the work you need done, then compare the operating commitment and evidence of success.
See how ai-coustics tested background noise across eight STT engines, then use Coval to test competing speech and other audio risks. Explore the results.
How our code reviewer came to run cheap models several times instead of a strong model once.
A Coval x Gradium case study: counting leading silence changed time to first audio scores, and made TTS faster. See the benchmark.
Coval and Phonely are giving voice agents a continuous feedback loop, closing the gap between production performance and voice agent improvement.
A four-stage field guide for choosing a voice provider with independent benchmarks, use-case tests, simulation, and production evidence.
Latency is what separates a natural conversation from a phone tree. TTFA, Coval's streaming methodology, and July 2026 results. Read the benchmark.
How to use Word Error Rate as an accuracy filter, interpret Coval's July 2026 STT snapshot, and choose among models that clear your production threshold.