Three products, built to check each other.
Lucia Operator OS is the intelligence operators talk to. Lucia Guest Concierge is the one guests meet — live at our founding client, Villa Valentín. Evaluation Labs decides whether both are good enough to trust. Switch between them below.
Lucia Operator OS — operational intelligence for hospitality
Running a property is heavy: tangled schedules, a crowded lobby, thin margins. The industry's answer has been more screens, more alerts, more software — all demanding attention from people already stretched thin. Lucia takes the opposite path. Not another login. Not another feed. A calm partner that reads the moment in front of an operator and returns a clear next move.
What Lucia is built around
- Intent recognition — understanding what the operator actually needs, not just what they typed.
- Operational prioritization — surfacing the next action that matters, with less cognitive load.
- Emotional containment — warm but not mushy; steady when the situation is stressful.
- Trust-state discipline — no overclaiming; saying what is and isn't known.
What "better" means for Lucia
Not prettier. Not longer. Not more confident. Better — measured against the human situation in front of her:
- More accurate intent recognition and clearer next actions.
- Fewer trust-state mistakes and less overclaiming.
- Better hand-holding under pressure, and continuity across refinements.
Lucia Guest Concierge — the Lucia guests actually meet
Most guests will never see a dashboard — they simply talk to Lucia. The Guest Concierge is the public, guest-facing surface of the same intelligence: hospitality conversations, concierge support, and property knowledge, delivered warmly and truthfully, with nothing to download.
What the concierge handles
- Guest conversations — questions, requests, and recommendations in natural language.
- Property intelligence — answers grounded in what is actually true for that property and that stay.
- Calm escalation — knowing when a human needs to step in, and handing off cleanly.
- The same trust rules — no overclaiming to guests, ever; the doctrine applies on both surfaces.
Live with our founding client
The Guest Concierge is official and operating — in live testing at Villa Valentín, our founding client property. Every conversation feeds the same evaluation loop in Eval Labs that hardens the rest of the family.
Evaluation Labs — how we teach Lucia to be trustworthy
Eval Labs is proprietary software we built from scratch: the evaluation, analysis, and human-judgment system used to test, refine, and protect Lucia's intelligence as it develops. It captures how Lucia performs under real operational pressure — and turns that into reliable evidence, not a pile of scores.
What it captures
- Prompts, Lucia's responses, and full run state.
- Human review, scores, notes, and final run lifecycle.
- Suggested selections the reviewer can accept, override, or escalate.
The core loop
This loop is the entire point. One good answer proves almost nothing; a repeated pattern proves a lot.
- Create or load a prompt suite
- Run Lucia’s responses
- Review & score each item
- Capture human guidance
- Finalize & export
- Identify a behavioral pattern
- Patch engine / prompt / routing
- Re-run the same suite & compare
Roles & current status
Eval Labs uses three roles — owner, admin, and evaluator. The evaluator role is intentionally narrow: run assigned custom evals, review and finalize your own runs, nothing wider.
- AI-reviewed platform-readiness gate passed: 60 runs · 3,000 prompts · 3,000 responses · 3,000 reviews.
- That proves platform readiness — not human approval of Lucia's quality.
- Human evaluation is the decisive Lucia-quality layer, and it's where researchers contribute.
Why three, not one
A product that grades itself isn't trustworthy. Two Lucia surfaces serve humans — operators running the property and guests staying in it — and Evaluation Labs exists as a separate product to judge both, with human reviewers inside the loop. That separation is the whole design. It's also what makes the evaluator role real, mentored work rather than a demo.