What it does
CallLens transcribes every call while it is still running, attributes each line to agent or customer, redacts personal data before anything is stored, and turns the transcript into a summary, action items, topics and a per-turn sentiment score that the agent confirms with one click. Because it runs on your own GPU, cloud transcription services and their data-protection questions are off the table.
What you get
- The speech model fine-tuned for German customer calls, with the streaming inference and diarisation service.
- The redaction stage for phone numbers, addresses, IBANs and card numbers, and the summarisation model.
- The agent view with live transcript and CRM hand-off, and the team-lead dashboard with sentiment trends per topic.
- A fine-tuning kit for your own consented audio, documentation, an onboarding session and a direct line to the engineers.
Deployment and licensing
One mid-range GPU keeps up with a centre handling well over a thousand calls a day. The one-time licence is perpetual, runs inside your network and includes twelve months of updates; the subscription adds model updates as they ship and can be cancelled monthly. Licensed per agent tier; telephony integration is scoped together with you.
Proven in production
In the service centre it was built for, the fine-tuned model reaches 6.1 % word error rate, down from 14.3 % for the base model, and after-call work dropped by around 40 %. The case study linked above has the details.