AI Assistants & Voice Agents
Conversational AI grounded in your knowledge base — voice, WhatsApp and web.
We build production-grade conversational AI systems: real-time voice agents (speech-to-text, LLM with RAG, text-to-speech under 200ms), automated document ingestion into a searchable knowledge base, and multi-channel delivery — interactive kiosks, WhatsApp, web portals and email — all on a single backend. Engineered against hallucinations with multi-layer grounding safeguards and built GDPR-first for public-sector deployment.
- Real-time voice pipeline: STT → RAG → LLM → TTS, Italian-native
- RAG on PostgreSQL/pgvector with multilingual embeddings
- Automated PDF ingestion: official documents parsed into the KB in minutes
- Multi-channel: voice kiosks, WhatsApp, web portal, email notifications
- Multi-layer hallucination prevention (KB-only grounding, NLI judge, auto-flagging)
- GDPR-first: k-anonymous analytics, retention purging, per-channel identity silos
- 3D avatar integration (Unreal Engine MetaHuman) with emotion detection
Case study
AI recruitment assistant for a national public institution
A voice and text AI assistant guiding citizens through enrollment procedures: automated ingestion of official tenders, eligibility matching, voice kiosks at live events, WhatsApp and web portal — with privacy-by-design analytics.
→ In production on kiosks, WhatsApp and web — answers under 200ms