Release Notes: Smart Kiosk Assistant#

2026.2.0#

This release delivers a unified Smart Kiosk Assistant stack with dual React-based UIs, OpenVINO-powered AI inference, agentic tool-calling ordering, queue-aware recommendations, enhanced voice interaction, optional multimodal identity, and streamlined deployment. This update includes the following changes:

  • The kiosk front end has been rebuilt as a React (Vite + TypeScript) single-page application, replacing the previous Gradio interface for a faster and more customizable web experience.

  • A single kiosk-ui image now serves two runtime modes selected by KIOSK_UI_MODE: an operator screen on port 7860 and a customer-facing kiosk screen on port 7861.

  • The operator screen provides voice and text chat, a live queue feed, knowledge-base ingestion, audio device settings, and a performance dashboard with executive KPIs, model KPIs, hardware charts, and pipeline flow visualization.

  • The customer screen provides a queue-aware menu grid with a category rail, a live cart panel with running order total and upsell prompts, a queue-status bar, and a voice “Ask” bar.

  • The ordering agent now drives the cart through an MCP tool server hosted by kiosk-core, with tools for catalogue browsing, cart lifecycle management, order confirmation, and upsell suggestions. Built-in guardrails resolve ambiguous item references and reject implausible quantities.

  • A rule-based upsell engine produces contextual add-on suggestions that are surfaced in both the cart and the spoken response.

  • The ordering agent runs its Qwen3-4B language model through OpenVINO Model Server (OVMS) instead of an in-process OpenVINO model, providing a dedicated, OpenAI-compatible inference endpoint for tool calling.

  • A queue analytics capability adds a person-counting service and an RTSP streamer, using YOLO detection with OpenVINO to track queue length from a video feed, expose a live MJPEG overlay stream, and surface a dynamic peak-hour menu.

  • The audio-analyzer service adds OpenAI-compatible streaming transcription over Server-Sent Events, a realtime WebSocket transcription endpoint, Video Summarization Service (VSS) response compatibility, and more accurate multi-speaker segment splitting with persisted enrolment.

  • Speaker diarization has been enabled across the audio-analyzer and kiosk-core pipeline, improving turn attribution during multi-speaker interactions.

  • The text-to-speech service now supports named voices and produces faster, more natural-sounding prosody.

  • An optional multimodal identity service adds Face ID and voiceprint authentication, combining OpenVINO face and ECAPA voice inference with a FAISS index and SQLite loyalty profiles, with login, registration, and enrolment screens in the UI, enabled through a dedicated deployment profile.

  • Inference devices are configurable per service (CPU, GPU, NPU), including NPU passthrough for identity-service, ovms-llm, and audio-analyzer ASR, with independent device settings for the RAG embedding and reranker models.

  • All kiosk services and container images have been rebuilt and aligned to 2026.2.0: kiosk-core, kiosk-ui, queue-service, identity-service, rag-service, rtsp-streamer, audio-analyzer, text-to-speech, and metrics-collector.

  • A Makefile-based workflow simplifies setup and operations with targets for environment initialization, configuration validation, model and sample-video download, image build, service startup, per-service health checks, log tailing, single-service rebuilds, and cleanup.

  • Sample-video tooling downloads and provisions the RTSP feed clips used by the queue analytics pipeline, configurable through the environment file.

2026.1.0#

The initial release of Smart Kiosk Assistant marks the launch of a voice-enabled interactive application for retail, QSR, Airlines and other customer-facing environments. The application has the following features:

  • Designed as a conversational AI experience, it enables users to engage naturally through speech and receive intelligent, spoken responses in real time.

  • The platform brings together speech recognition, retrieval-augmented generation, and text-to-speech in a seamless, end-to-end voice interaction flow.

  • With browser-based voice capture and natural audio playback, the experience feels intuitive, responsive, and ready for real-world engagement.

  • Smart Kiosk Assistant grounds every response in an ingestible local knowledge base, helping deliver more relevant, context-aware, and business-specific answers.

  • Its integrated AI stack combines kiosk UI, orchestration, speech-to-text, retrieval, and speech synthesis into a unified deployment-ready application.

  • The experience is further enhanced by built-in visibility into model KPIs and live performance data, including runtime model details and latency metrics.

  • Optimized for local and edge deployment, the application leverages OpenVINO acceleration on Intel hardware for efficient AI inference.

  • Docker Compose packaging and flexible configuration make the solution easy to deploy, adapt, and scale across enterprise environments.

  • This launch establishes Smart Kiosk Assistant as a strong foundation for immersive, intelligent, and voice-first digital engagement experiences.

On this page

This Page