AI Call Analytics
Turn every recorded call into a transcript, a QA score, and an answer — automatically.
Astervis listens to your calls so your supervisors don't have to. Every recorded conversation can be transcribed, analysed, scored against your own quality rules, and made searchable — without anyone pressing play.
This is the headline of Astervis v1. It runs on top of the same call data you're already collecting, so there's nothing new to deploy on your PBX — you just turn it on.
AI analysis is optional and metered. The analytics platform works fully without it. When you enable it, only the audio of a finished call is sent to Google Gemini for analysis — see AI & Privacy for exactly what leaves your server and what doesn't.
What you get from every call
Transcription & diarization
Full text with each turn labelled agent or customer, in the original language.
Sentiment & category
Positive / neutral / negative, plus sales, support, complaint, inquiry and more.
QA scoring
A 0–100 score against your own configurable rubric, with reasons and quotes.
Semantic search & assistant
Ask questions in plain language and search across what was actually said.
Agents & insights
Daily coaching cards and autonomous monitors that watch your metrics for you.
How it works
When a call ends and its recording lands on the server, the on-prem queue submits the audio to the AI gateway. A single multimodal Gemini call (on Google Vertex AI) does transcription, speaker separation, analysis, and — if you use it — rubric-based QA, all at once. The result is written back into your database, ready on the call card.
A call finishes
Astervis already replicates the call record. If a recording exists, it becomes eligible for analysis.
The audio is analysed
The queue sends the recording to Gemini. Status moves
pending → processing → completed. Rate-limited jobs back off and retry on their own.
Results appear on the call
Transcript, summary, sentiment, category, topics, key quotes, recommended next actions and a QA score show up on the call detail view and feed the dashboards.
Built for ru / uz / en
Russian, Uzbek and English are first-class — prompts are written in all three, and summaries come back in Russian regardless of the call language. Kazakh, Kyrgyz, Turkish, Arabic, Tajik and Persian are recognised downstream too.
Enabling AI
AI is gated by a separate AI license tier (it has its own rate limit and monthly quota, and a flag for QA scoring). If your plan includes it, analysis starts running automatically on new calls. To score historical calls under a new rubric, an operator can run a one-off rescore — see QA scoring.
Last updated on