HiJiffy Aplysia 3 (guest-facing hotel AI, Console co-pilot and insights)

par HiJiffy

AI Application

Faible confiance

Aplysia 3 is HiJiffy's proprietary AI model for hotels. It answers guest messages automatically across webchat, Facebook Messenger, Instagram, WhatsApp, Telegram, voice clips and OTA messaging (Booking.com, Expedia, Airbnb), can walk guests through chat-based bookings, drafts email replies for human review, and acts as a co-pilot in the staff Console (translation, tone and grammar polish, suggested replies, pre-filled booking details). It reads PMS and booking-engine context such as arrival dates and stay status, detects intent and sentiment, runs proactive campaigns, and reports knowledge gaps. The vendor describes a stack of machine learning, deep learning, NLP, retrieval-augmented generation over a hotel's verified documents, unnamed large language models, embeddings, speech-to-text, memory and guardrails in 130+ languages, and says it handles up to 90% of conversations without a human in the loop, with per-source visibility for each answer and escalation when confidence is low. Rated medium: it is a guest-facing AI handling guests' personal data at scale with limited human review, and the vendor documents guardrails, per-source reporting, a sub-processor list, EU hosting (AWS Ireland) and a 2025 penetration-test statement. Open points: no SOC 2, ISO 27001 or similar certification is stated for HiJiffy itself; the AI page says the model learns from conversations and staff corrections without saying whether one customer's data improves models used for others; the large language models producing replies are not named (OpenAI is listed only 'for AI add-ons'); no accuracy or escalation-error rates are published; and the customer agreement, DPA and AI terms are not public. The rating would rise if cross-customer learning is confirmed without opt-out, or if guest data is processed outside the EU by unnamed model providers.