Case study

Sonar — real-time video advisory for German banks

Client / Context

Sonar is a European video-consultation platform whose users include German banks. Vaagmodo is Sonar's long-term engineering partner, contributing continuously across the platform's real-time video, AI and integration work.

Banking is a maximum-scrutiny environment: European data-protection law (GDPR/DSGVO), high availability expectations and conservative IT governance. Everything delivered into a platform like this has to hold up to that standard, year after year.

Challenge

  • Deliver and continuously evolve real-time video consultation that works reliably across browsers, devices and imperfect networks — for advisors and their customers.
  • Add live multilingual captioning and translation so consultations work across language barriers — without routing sensitive audio through third-party cloud AI services.
  • Sustain this over years as a stable engineering partnership, not a one-off project.

Solution

  • Real-time video engineering — WebRTC and server-side media work delivered as part of the platform's continuous development: making calls connect, record and stay usable across browsers, devices and networks nobody controls.
  • Live captioning and translation in 15 languages (including Indian languages), integrated into video consultations: Whisper ASR for speech recognition and self-hosted open translation models, running on infrastructure the operator controls, so audio never leaves it.
  • AI in the video path — computer vision (a self-hosted open vision-language model) for damage analysis in insurance video calls, assessing damage shown on camera during a live call. It began as a small feasibility prototype, which is how we prefer to start anything uncertain.
  • Payment and checkout integrations delivered inside the same platform context — SEPA and card flows.
  • Operations discipline — containerized deployment, CI/CD and monitoring as standing practice.

Years of production real-time media work is also what equips us to migrate legacy video platforms to modern stacks (LiveKit, mediasoup-class SFUs) — the architecture we default to for new builds. The engineering behind self-hosted AI capacity is covered in our AI Infrastructure case study.

Outcome

  • Vaagmodo-engineered technology runs in video consultation used by German banks — a named, referenceable client in a regulated sector.
  • Live captioning and translation across 15 languages including Indian languages, delivered GDPR-compatibly through fully self-hosted AI, with no third-party AI API in the data path.
  • Computer vision added to the live video path for insurance damage assessment, from feasibility prototype to production feature.
  • A long-term engineering partnership that continues today — our strongest evidence of delivery quality.

Reference architecture

Advisor browser (WebRTC) Customer browser (WebRTC) Signaling & media negotiation Server-side media layer RTP routing · recording · in-pipeline processing Speech recognition open ASR models Translation open models, 15 languages Self-hosted GPU infrastructure — audio never leaves it live captions back into the consultation audio/video
An illustrative reference architecture for video consultation with live captioning — how we build this class of system. It is not a depiction of any client's production system.
languages in live captioning and translation, including Indian languages
15 languages in live captioning and translation, including Indian languages
compatible by architecture — fully self-hosted AI in the data path
GDPR compatible by architecture — fully self-hosted AI in the data path
client reference in the German banking sector, available on request
Named client reference in the German banking sector, available on request

This case study is structured for use as a bid annexure. A client reference can be arranged on request via info@vaagmodo.com, subject to client consent.

Building a video platform for a regulated industry?