In buildEst. Q4 2026evalsvoicegovernance

voice-eval

Evaluation harness for AI voice agents. Because 'sounds fine on the demo call' isn't a strategy.

voice-eval preview

§ 01 · What it is

An open-source evaluation harness that scores every call an AI voice agent handles — for latency, intent accuracy, escalation correctness, cost per call, and PII leakage. Point it at a recording, feed it the expected metadata, and get a pass/fail with actionable failure breakdowns. Built to catch drift before your customers notice.

§ 02 · Money it saves

£8,000-£30,000/yr per voice agent
for any business running an AI voice agent (hospitality, professional services, healthcare, trades)
The math: A single bad call caught before it goes live prevents ~£300-£800 of downstream refund/goodwill/reputation cost. A drift regression caught in a nightly eval run instead of by a customer complaint prevents ~£2,000-£5,000. Most production voice agents currently ship without any of this — the £8-30k range is the cost of one bad month of undetected quality decay.

§ 03 · How it works

  1. 01

    Records every call with metadata: latency at each stage, model version, prompt version, confidence scores, cost, escalation decisions.

  2. 02

    Compares each call against a rubric — pass thresholds for latency (< 1.5s), intent accuracy (> 95%), escalation correctness (100% for high-value bookings), and hallucination detection (LLM-as-judge scoring against grounded data).

  3. 03

    Runs nightly against a held-out set of 20-50 golden calls so model updates or prompt changes get caught before they hit production.

  4. 04

    Ships with a CLI + web dashboard. Exports Slack alerts when quality drops.

Stack

TypeScriptNodeOpenAI Whisper (grounding)Claude (LLM-as-judge)SQLiteCLI + Next.js dashboard

§ 04 · Why it beats what's out there

Every general-purpose LLM eval framework treats voice as an afterthought — you feed it text, it scores text. Voice has specific failure modes (per-stage latency, escalation correctness, PII in transcripts) that text-first tools miss.

What's out thereThe gap it leaves
General-purpose LLM eval platformsBuilt text-first. Don't handle latency-at-each-pipeline-stage, escalation correctness, or PII leakage in voice transcripts — the three failure modes that actually matter for a voice agent in production.
Analytics baked into commercial voice-agent platformsVendor lock-in. If you switch your voice stack, you rebuild your evals from scratch. This runs against any provider — swap models or vendors freely and your evaluation history moves with you.
Manual review of random call samplesDoesn't scale past a few hundred calls a month. Catches quality drift after customers do, not before. Nightly automated eval runs catch drift the same night.
Want this in your business?

I'm building this now. If it fits your problem, let's talk before it ships.

Follow along

Build notes and progress on this system will be posted in the Notes. Or check GitHub when the repo goes public.