Milestonev0.4 records the moment Lucia moved from phrase-heavy routing toward a hybrid intent system: deterministic safety plus semantic model assistance, confidence arbitration, runtime evidence, and Eval Labs proof.
Why this milestone matters
Lucia’s long-term product advantage is not a prettier dashboard or a nicer chatbot voice. The emerging moat is Lucia’s ability to understand what the operator means inside real hospitality operations while preserving calm, truth, and boundaries. The hybrid intent layer is the first runtime proof of that architecture.What changed
Lucia can now interpret selected ambiguous or compressed language through a bounded semantic assist layer. Examples validated in evals:- “Let’s do this” after Lucia offers a top item / rundown
- “All right, let’s begin” after a social bridge
- “Start me there” as top-item acceptance
- “Take me into it” as a continuation request
- “Good night” / “Sleep well” / “See you tomorrow” as safe human closings
- weather prompts as property-context utility
Weather-context utility
Weather is not generic off-topic anymore because it affects hospitality operations. Lucia now clarifies whether weather should be anchored to:- the user’s current location
- Villa Valentin
- future managed properties
Validation proof
- Strict eval:
47/47 - Runtime revision: recorded in the owning source history
- core operational routing
- social/human utility
- low-risk hospitality humor
- bounded off-role behavior
- semantic follow-up acceptance
- valediction handling
- weather-context clarification
- delayed follow-up expiry
Strategic meaning
The proprietary layer is not the base model alone. The proprietary layer is the system around it:- intent ontology
- routing policy
- arbitration rules
- confidence thresholds
- eval bank
- operational state model
- Villa Valentin operating data
- emotional containment doctrine
- observability tooling