VICIPanel
Telephony Intelligence · AMD

AMD that gets smarter every week you run it

Stock VICIdial AMD hasn't changed meaningfully since 2008. VICIPanel wraps four detection layers around the same engine, acoustic fingerprinting, native AMD, post-call AI verification, and a real-time AGI classifier, and every confirmed machine feeds back into the fingerprint database.

Detection layers

4

Fingerprint signal

MFCC acoustic

Self-learning

Per campaign

What it does

Four-layer AMD with self-learning

Most dialers rely on silence-duration heuristics baked into Asterisk over a decade ago. They misclassify live humans as voicemails, and voicemails as live humans, at rates that cost you real money. VICIPanel keeps the native layer for speed but stacks three more layers on top: MFCC acoustic fingerprints that recognize the specific voicemail greetings common to each carrier, post-call AI verification that reviews what was actually said, and a real-time AGI classifier that makes routing decisions in milliseconds. Every call that gets corrected, either by your agent dispositioning or by the AI verification layer, feeding back into the fingerprint database. Your AMD gets more accurate every week.

Answering Machine Detection — captured from the live demo

How it works

Answering Machine Detection in 4 steps

Layer 0

Acoustic fingerprint (MFCC)

First audio frames get converted to Mel-Frequency Cepstral Coefficients and matched against a per-campaign fingerprint database. Known voicemail greetings get flagged instantly.

Layer 1

Native Asterisk AMD

The standard silence-based AMD runs in parallel as a baseline, useful for unfingerprinted numbers and as a fallback when Layer 0 is uncertain.

Layer 2

Real-time AGI classifier

An AGI process makes the actual routing decision combining the signals from Layers 0 and 1, with configurable confidence thresholds per campaign.

Layer 3

Post-call AI verification

After the call ends, Whisper transcribes the first seconds and an LLM verifies the classification. Corrections feed back into the fingerprint database as training data.

Highlights

What makes this different

Per-campaign fingerprint tuning

Different campaigns call different audiences on different carriers. The fingerprint database is scoped per campaign so your solar dial-list doesn't pollute your healthcare list.

Transparent decision log

AMD decisions are logged with the triggering layer and confidence score. Supervisors can audit AMD decisions per call without digging through Asterisk logs.

Works on your existing VICIdial

No changes to your Asterisk config. VICIPanel drops in next to the native AMD and augments it, roll back in an afternoon if you don't like it.

Why it's cutting edge

What VICIPanel answering machine detection does differently

Native VICIdial AMD hasn't been meaningfully updated in over a decade, we augment it instead of forking it

Self-learning database means accuracy compounds week over week, without manual tuning

Four layers with explicit confidence thresholds instead of a single opaque algorithm

Every decision is audit-logged so compliance teams can defend any routing choice

See it running on your VICIdial

The live demo runs on real VICIdial data. Try everything above — instant credentials, no signup, no sales call.