A voice assistant you operate yourself

Five-phase architecture

Every citizen call runs through the same five phases — Listening, Understanding, Deciding, Protecting, Connecting. Detailed phase descriptions are on the home page in the hero section.

Core capabilities

Eleven production-ready building blocks, composable per use case:

• Real-time Language ID

  Identifies the caller's spoken language within the first 3 seconds of speech using an acoustic fingerprint model, enabling seamless multilingual routing without requiring the caller to select a language.

• Bayesian Comprehension Fusion

  Combines ASR confidence scores, dialogue-act probabilities, and domain-specific priors in a Bayesian network to produce a calibrated comprehension score — triggering agent handoff only when understanding drops below a configurable threshold.

• Fully Self-Hosted

  Runs entirely on your own infrastructure — on-premises server room or sovereign-cloud VM — with zero external API calls; no citizen voice data ever leaves your network perimeter.

• Whisper ASR

  Leverages OpenAI Whisper (large-v3-turbo) fine-tuned on German administrative vocabulary and regional dialect data for best-in-class word-error rates on citizen-service call recordings.

• Piper TTS

  Generates natural-sounding German, Turkish, Arabic, French, Russian, and English voice responses using Piper, a fast neural text-to-speech engine that runs in real time on CPU with under 150 ms latency.

• Qdrant Vector RAG

  Retrieves authoritative answers from your citizen-service knowledge base — forms, deadlines, eligibility rules — via semantic vector search in a local Qdrant instance, grounding every response in up-to-date official documents.

• Immutable Audit Chain

  Writes a cryptographically linked, append-only log of every call event — ASR transcript, intent classification, handoff decision, operator action — enabling full reproducibility for supervisory authority audits under the EU AI Act.

• Eco-Metrics Dashboard

  Measures and reports real-time energy consumption per call (kWh), total CO₂-equivalent footprint, and comparative savings versus cloud-based alternatives to support your authority's sustainability reporting obligations.

• V-Modell XT Compliant

  Developed and documented in accordance with V-Modell XT, the mandatory German public-sector software development standard, including traceable requirements, defined test levels, and a structured handover package.

• GDPR Shield

  Built-in technical and organisational measures (TOMs) satisfy GDPR Art. 25 (privacy by design), Art. 32 (security of processing), and Art. 35 (DPIA) — with pre-filled records-of-processing templates included.

• Smart Handoff UX

  When confidence drops below threshold, the system hands off to a human agent with a one-screen context card: live transcript, detected intent, caller language, and suggested response — reducing average handling time by up to 40%.

Compliance

• GDPR Art. 25 — privacy by design, no external data flows
• GDPR Art. 32 — security of processing, end-to-end encryption of operator sessions
• GDPR Art. 35 — DPIA template included in the delivery package
• EU AI Act — high-risk classification (Annex III §5), full technical documentation
• V-Modell XT — traceable requirements, defined test levels, structured handover
• EN 301 549 / WCAG 2.1 AA — operator UI meets public-sector accessibility requirements

Integration

VoiceA integrates via SIP trunk or WebRTC with your existing telephony and can index knowledge bases in common formats (Markdown, PDF, DOCX) as well as from domain systems (for example via ELAK or registration-law interfaces).